Download - Reinhardt Bernhard - LMU
On the Retrieval of CircumsolarRadiation from Satellite Observations
and Weather Model Output
Dipl.-Met. Bernhard Ulrich Reinhardt
Munchen 2013
On the Retrieval of CircumsolarRadiation from Satellite Observations
and Weather Model Output
Dipl.-Met. Bernhard Ulrich Reinhardt
Dissertation
der Fakultat fur Physik
der Ludwig-Maximilians-Universitat
Munchen
vorgelegt von
Dipl.-Met. Bernhard Ulrich Reinhardt
Institut fur Physik der Atmosphare
Deutsches Zentrum fur Luft- und Raumfahrt
Oberpfaffenhofen
Munchen, August 2013
Erstgutachter: Prof. Dr. Bernhard Mayer
Zweitgutachter: Prof. Dr. Markus Rapp
Tag der mundlichen Prufung: 20.11.2013
Contents
Zusammenfassung vii
Abstract ix
1 Introduction 11.1 Circumsolar Radiation and Concentrating Solar Technologies . . . . . . . . 11.2 Aims of this Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Basic Principles 72.1 The Radiative Transfer Equation . . . . . . . . . . . . . . . . . . . . . . . 72.2 Scattering of Sunlight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.3 Remote Sensing of Clouds from MSG . . . . . . . . . . . . . . . . . . . . . 12
2.3.1 The SEVIRI Instrument . . . . . . . . . . . . . . . . . . . . . . . . 122.3.2 Retrieval of Cloud Properties from Reflected Solar Radiation . . . . 15
2.4 Passive Satellite Remote Sensing of Aerosol . . . . . . . . . . . . . . . . . 172.5 The Sunshape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192.6 The Circumsolar Ratio – CSR . . . . . . . . . . . . . . . . . . . . . . . . . 212.7 Ground Based Measurements of the Circumsolar Radiation . . . . . . . . . 23
3 Tools and Methods 253.1 Optical and Microphysical Properties of Cirrus Clouds . . . . . . . . . . . 253.2 Cloud Property Remote Sensing with the APICS Algorithm . . . . . . . . 27
3.2.1 Algorithm Description . . . . . . . . . . . . . . . . . . . . . . . . . 273.2.2 Auxiliary Albedo Dataset . . . . . . . . . . . . . . . . . . . . . . . 283.2.3 Cirrus Cloud Mask . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3 Aerosol Input and Processing . . . . . . . . . . . . . . . . . . . . . . . . . 333.3.1 Modeled Aerosol Data from the ECMWF IFS . . . . . . . . . . . . 333.3.2 Aerosol Optical Properties . . . . . . . . . . . . . . . . . . . . . . . 343.3.3 Post-Processing of ECMWF Output . . . . . . . . . . . . . . . . . 39
3.4 Atmospheric Radiative Transfer . . . . . . . . . . . . . . . . . . . . . . . . 403.4.1 The RTE-Solvers DISORT and MYSTIC . . . . . . . . . . . . . . . 403.4.2 Adaptation of the RTE-Solver MYSTIC . . . . . . . . . . . . . . . 423.4.3 Pseudo-Spectral and Solar Integrated Radiative Transfer . . . . . . 50
vi Inhaltsverzeichnis
3.5 Parametrization of Circumsolar Radiation . . . . . . . . . . . . . . . . . . 523.5.1 Concept of the Apparent Optical Thickness . . . . . . . . . . . . . 533.5.2 Treatment of Multiple Scattering Layers – The Adding Method . . 583.5.3 Sensitivity of the CSR Parameterization on Sun Zenith Angle . . . 643.5.4 Sensitivity of the CSR Parameterization on the Scattering Layer’s
Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 653.5.5 Further Aspects of the Apparent Optical Thickness Approach . . . 66
3.6 Sensitivity of Circumsolar Radiation to Ice Particle Shape . . . . . . . . . 673.7 Sensitivity of CSR to Aerosol Particle Shape and Relative Humidity . . . . 68
4 Results 734.1 Intercomparison of Retrieval Results for Different Ice Particle Shapes . . . 734.2 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.2.1 Validation of Aerosol Related Circumsolar Radiation . . . . . . . . 794.2.2 Validation of Cirrus Related Circumsolar Radiation . . . . . . . . . 82
4.3 Characteristics of Cirrus-Related Circumsolar Radiation . . . . . . . . . . 894.3.1 Spatial Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 894.3.2 Irradiance Overestimation by Pyrheliometers . . . . . . . . . . . . . 914.3.3 Relation of Irradiance and CSR . . . . . . . . . . . . . . . . . . . . 94
5 Discussion 975.1 Statistical Post-Processing of IFS Aerosol Output . . . . . . . . . . . . . . 975.2 Uncertainties in the Cirrus Cloud Property Retrieval . . . . . . . . . . . . 100
5.2.1 Undetected Cirrus Clouds . . . . . . . . . . . . . . . . . . . . . . . 1015.2.2 Spurious SEVIRI Pixels . . . . . . . . . . . . . . . . . . . . . . . . 107
5.3 Comparison of the Satellite CSR Retrieval to a Regression Model . . . . . 107
6 Conclusions 1116.1 Synopsis of Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1126.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Bibliography 115
Glossary 125
Acknowledgements 127
Zusammenfassung
Konzentrierende Solarkraft steht im Wettstreit mit anderen sich dynamisch entwickel-
nden erneuerbaren Energiequellen. Ein zentraler Erfolgsfaktor ist dabei die Verringerung
der Energieerzeugungskosten. Bei der Verfolgung dieses Ziels kommt der verbesserten
Bestimmung der Einstrahlung eine wichtige Rolle zu. Finanzierungskosten fur Neuan-
lagen konnen durch eine genauere Vorhersage der solaren Resource verringert werden,
weil diese zu einer Reduzierung der finanziellen Risiken fuhrt. Desweiteren erlaubt eine
verbesserte Bestimmung der Einstrahlung zukunftige Anlagen in Bezug auf die lokalen
Bedingungen zu optimieren. Dies senkt die Kosten und erhoht die Energieeffizienz. Die
Zirkumsolarstrahlung ist ein Parameter, dem bei der Resourcenbestimmung immer mehr
Aufmerksamkeit zu Teil wird. Sie wird durch Vorwartsstreuung des Sonnenlichts an
Wolken- und Aerosolpartikeln hervorgerufen. Die Messung von Zirkumsolarstrahlung ist je-
doch anspruchsvoll und es existieren nur wenige Messreihen von eingeschranktem Umfang.
Um die Lucke zwischen der vermehrten Nachfrage nach Daten zur Zirkumsolarstrahlung
und deren eingeschrankter Verfugbarkeit zu fullen, wurde in dieser Arbeit eine Meth-
ode zur Bestimmung der Zirkumsolarstrahlung aus verfugbaren Datensatzen von Aerosol-
und Wolkeneigenschaften entwickelt. Im Speziellen wurden die optische Dicke und der
Effektivradius von Cirrus-Wolken, sowie die Flachenmassenkonzentration verschiedener
Aerosolkomponenten ausgewertet. Den Kern der Methode zur Ableitung der Zirkumso-
larstrahlung stellt eine schnelle und dennoch genaue Parametrisierung dar. Diese erlaubt
es die Zirkumsolarstrahlung mittels einfacher analytischer Ausdrucke aus zuvor tabel-
lierten Koeffizienten zu berechnen, anstatt den Strahlungstransport zeitaufwandig nu-
merisch zu berechnen. Die entsprechenden Tabellen wurden mittels umfangreicher Simula-
tionen mit einer speziell angepassten Version des Monte Carlo Strahlungstransportmodells
MYSTIC erstellt. MYSTIC wurde im Rahmen der Studie unter anderem um eine realistis-
che Strahlungsquelle erweitert, indem die bisher verwendete Punktquelle durch eine aus-
gedehnte Sonnenscheibe mit wellenlangenabhangiger Helligkeitsverteilung ersetzt wurde.
Die ausgewerteten Aerosolflachenmassenkonzentrationen wurden als Modellausgabe des
Integrated Forecast System (IFS) vom European Centre for Medium-Range Weather Fore-
casts (ECMWF) bezogen. Um die Wolkeneigenschaften von Cirren abzuleiten wurde das
APICS Retrievalsystem auf Messungen der Meteosat Second Generation (MSG) Satelliten
angewandt. Im Zuge der Studie wurde APICS fur das Retrieval optisch dunner Cirren
optimiert. Dazu wurde ein neuer Datensatz der Bodenalbedo auf Basis von MSG Messun-
gen generiert, der als a priori Annahme in das Retrieval einfließt. Dieser neue Datensatz
viii Zusammenfassung
ist, im Gegensatz zu dem bisher verwendeten, konsistent zu den anderen innerhalb des
Retrievals getroffenen Annahmen. Dies ist eine wichtige Voraussetzung fur die Ableitung
der Eigenschaften von optisch dunnen Cirren. Desweiteren wurde APICS mit einer neuen
Wolkenmaske betrieben, die auf der Ausgabe des COCS-Cirrenretrievals basiert. Sie ersetzt
die zuvor genutzte Wolkenmaske des MeCiDa Eiswolkendetektionsalgorithmus. Dadurch
konnen etwa 70 bis 80 Prozent mehr von jenen dunnen Cirren berucksichtigt werden, die
noch genug Sonnenlicht zum Betrieb einer typischen solar-thermischen Anlage passieren
lassen.
In Bezug auf Cirren stellt die Form der Eispartikel einen Unsicherheitsfaktor dar –
sowohl bei der Ableitung der Wolkeneigenschaften, als auch bei der Berechnung der Zirkum-
solarstrahlung. Bisher gibt es noch keine Moglichkeit die Partikelform von MSG aus zu
bestimmen, sondern es muss eine Annahme a priori gemacht werden. Um eine Bestim-
mung der daraus folgenden Unsicherheit zu ermoglichen wurde APICS erweitert, so dass
mehrere neue Eispartikelformen bei der Ableitung der Wolkeneigenschaften verwendet wer-
den konnen. Damit konnte eine Unsicherheit in der Großenordnung von bis zu 50% in der
mittleren Zirkumsolarstrahlung festgestellt werden, die aus der Unbestimmtheit der Eis-
partikelform folgt.
Die entwickelte Methode zur Ableitung der Zirkumsolarstrahlung wurde mit Boden-
messungen des Zirkumsolarverhaltnis (engl. circumsolar ratio, CSR) validiert, die an
der Plataforma Solar de Almerıa (PSA) durchgefuhrt wurden. Dabei zeigte sich, dass
die statistische Verteilung der Zirkumsolarstrahlung mit beiden der verwendeten “Baum”
Eispartikelformmischungen gut charakterisiert werden kann. Beim Vergleich instantaner
Werte treten jedoch Timing- und Amplitudenfehlern auf. In der Validierung zeigte sich fur
das CSR eine mittlere absolute Abweichung (engl. mean absolute deviation, MAD) von
0.11 fur beide “Baum” Parametrisierungen, ein Bias von 4% bzw. -11% und eine Spear-
man Rang-Korrelation rrank,CSR von 0.54 bzw. 0.48. Wenn Messungen mit sub-skaligen
Cumulus Wolken innerhalb der entsprechenden Satellitenpixel manuell ausgefiltert wur-
den, verbesserte sich die Ubereinstimmung instantaner Werte. Dies spiegelt sich wider in
MAD-Werten von 0.08 bzw. 0.07 und rrank,CSR-Werten von 0.79 bzw. 0.76. Des Weit-
eren stellte sich heraus, dass das von Aerosol verursachte CSR deutlich unterschatzt wird,
wenn die Daten vom IFS unmodifiziert verwendet werden. Erst nach einer Anpassung der
Aerosolflachenmassenkonzentration konnen sinnvolle Ergebnisse erzielt werden. Vermut-
lich ist eine zu geringe Konzentration von großen Mineralstaubpartikeln im IFS der Grund
fur die Unterschatzung der CSR.
Die entwickelte Methode kann in Zukunft ausgeweitet und mit anderen Datenquellen
kombiniert werden. Wahrend bodengebundene Referenzmessungen bisher die Beurteilung
der Zirkumsolarstrahlung an nur wenigen Messstationen zulassen, konnen mit der hier neu
entwickelten Methode beliebige Kraftwerksstandorte begutachtet werden.
Abstract
Concentrating solar technologies compete with other rapidly developing renewable energy
sources. To succeed it is vital to lower the levelized cost of energy. There are several
parameters that can be optimized to reach this goal, but a key component is the im-
provement of the resource assessment. A better prediction of the solar resource for new
facilities brings down financing costs as financial risks are reduced. Moreover, improved
solar resource assessment allows to optimize new facilities in regard to the local insolation
conditions. This increases energy and cost efficiency. One parameter that is becoming
more and more important for the resource assessment is the circumsolar radiation. It is
caused by forward scattering of sun light by cloud or aerosol particles. However, measuring
circumsolar radiation is demanding and only very limited data sets are available. As a step
to bridge this gap, a method was developed in this study to determine circumsolar radi-
ation from readily available data on clouds and aerosol. Specifically, the effective radius
and optical thickness of cirrus clouds were used, as well as area mass loadings of several
aerosol components. The core of the method to determine the circumsolar radiation is a
fast yet precise parameterization. It allows to compute the circumsolar radiation by simple
analytical expressions from previously tabulated coefficients, instead of solving the radia-
tive transfer by time-consuming numerical simulations. The lookup tables were generated
by extensive calculations using a specifically adjusted version of the Monte Carlo radiative
transfer model MYSTIC. To this end, MYSTIC was enhanced with a realistic radiation
source: The point source used so far was replaced by a extended sun disk which features
a wavelength dependent brightness distribution.
The evaluated aerosol area mass loadings were obtained from the European Centre for
Medium-Range Weather Forecasts (ECMWF) as model output of the Integrated Forecast
System (IFS). To derive the cirrus cloud properties the APICS retrieval framework was
applied to Meteosat Second Generation (MSG) measurements. During the course of this
study APICS was optimized regarding the retrieval of optically thin cirrus clouds. To
this end, a new ground albedo data set was generated on the basis of MSG measurements
which serves as a priori assumption in the retrieval. This new data set is, in contrast
the so far used one, consistent to the other assumptions made within the retrieval. This
is an important pre-requisite for the successful retrieval of optically thin cirrus clouds.
Furthermore, APICS was operated with a new cloud mask based on output of the COCS
cirrus cloud property retrieval algorithm. It replaces the formerly used cloud mask from the
MeCiDa cirrus detection algorithm. Thereby in the order of 70 to 80 percent more optically
thin cirrus clouds can be considered, which allow enough light to pass for operation of a
x Abstract
typical solar thermal utility.
Considering cirrus clouds the prevailing ice particle shape is an uncertainty factor in
the cloud property retrieval as well as in the computation of circumsolar radiation. So
far it cannot be determined from MSG but must be assumed a priori. To allow for an
uncertainty analysis concerning this parameter APICS was extended to consider several
new ice particle shapes in the retrieval process. It was found, the nescience of the ice
particle shape leads to an uncertainty of up to 50% in the mean circumsolar irradiance.
The newly developed method for the retrieval of circumsolar radiation was validated
with ground measurements of the circumsolar ratio (CSR) performed at the Plataforma
Solar de Almerıa (PSA). This showed that the statistical distribution of the circumsolar
radiation can be well characterized with both of the two employed “Baum” ice particle
shape parameterizations. When comparing instantaneous values timing and amplitude
errors become evident, tough. For the circumsolar ratio (CSR) the validation yielded a
mean absolute deviation (MAD) of 0.11 for both “Baum” parameterizations, a bias of
4% and -11%, respectively, and a Spearman rank correlation rrank,CSR of 0.54 and 0.48,
respectively. If measurements with sub-scale cumulus clouds within the relevant satellite
pixels were manually removed, the agreement of instantaneous values improved. This
reflects in the MAD values of 0.08 and 0.07, respectively, and rrank,CSR values to 0.79
and 0.76, respectively. Furthermore, it was found that for aerosol the CSR is strongly
underestimated if the IFS output is used head on. Only after adjusting the aerosol mass
loadings reasonable values can be obtained. An underrepresentation of large dust particles
in the IFS seems most likely to be reason for this.
In the future the method developed in this study can be extended and combined with
other data sources. While ground-based reference measurements so far only allowed the
assessment of the circumsolar radiation at few specific measurement sites, the newly de-
veloped method makes it possible to survey arbitrary sites.
Chapter 1
Introduction
This thesis deals with the retrieval of circumsolar radiation to the benefit of concentrating
solar technology (CST). The term circumsolar indicates that the radiation is observable in
an annular region of the sky of a few degrees extent surrounding the sun. The phenomenon
of enhanced brightness in the circumsolar region is also referred to as aureole (comp.
Fig. 1.1).
1.1 Circumsolar Radiation and Concentrating Solar
Technologies
Concentrating solar technology denotes systems that concentrate the sun’s radiation to
harvest its energy. Concentration of the sun light is required if the natural non-concentrated
solar radiative flux density is not sufficient for the considered purpose – the most obvious
application is producing high temperatures. Concentration of radiation is inseparable from
a limiting of the angular region from which radiation can be utilized. This comes from
the fact that the etendue of a infinitesimal pencil of light dU = n dA dΩ is conserved
in geometric optics (Chaves, 2008). n is the refractive index of the medium, dA the
infinitesimal area perpendicular to propagation direction through which the light passes
and dΩ the solid angle covered by the pencil. CSTs typically have acceptance half-angles
α between 0.7 and 2.3 (Blanc, 2013, in regard to solar thermal systems). These systems
therefore require a tracking mechanism to follow the position of the sun. The tracking
comes with the advantage of the sun rays falling approximately perpendicular onto the
2 1. Introduction
system. No reduction of the incident irradiance through the so called “cosine effect” occurs,
as it is the case with non-tracking solar technologies, like common photovoltaics (e.g. King,
1996). However, concentrating systems cannot utilize most of the diffuse radiation from
the upper half space. The primary measure used to quantify the energy input into a
concentrating system is the direct normal irradiance (DNI), while for non-concentrating
systems the global horizontal irradiance (GHI) is of importance. The CSTs aperture sizes
may seem narrow but they still are considerably larger than the sun disk, which exhibits an
average angular radius αsun of 0.2661. Therefore it is somewhat ambiguous to use the direct
normal irradiance as quantity for the energy input. While in radiative transfer modelling
direct radiation is considered as radiation which has not been scattered, in experimental
meteorology it is defined as the radiation measured by a pyrheliometer, with α usually
2.5 (World Meteorological Organization standard according to CIMO Guide, chap. 7).
Certainly both definitions will not match the energy ingress through varying CST apertures
exactly. Indeed the deviations may be quite large. Furthermore not all radiation accepted
by a CST system is converted equally well. The conversion efficiency usually depends on
the angle of incidence. Both issues are connected to circumsolar radiation: Considering
pyrheliometer measurements, circumsolar radiation can cause an overestimation of the
solar irradiance available as resource for CST. Moreover, for the design and the precise
evaluation of high flux solar concentrators the radiance distribution inside and around the
sun disk, called sunshape, needs to be known (e.g. Rabl and Bendt (1982), Schubnell
(1992), Neumann and Witzke (1999) or Buie and Monger (2004)). To take these effects
into account, CST operators, constructors and designers need methods to determine the
circumsolar radiation entering different sized CST apertures.
Circumsolar radiation is caused by forward scattering of sun light by cloud or aerosol
particles. If these particles are horizontally evenly distributed, the radiance decreases with
angular distance from the sun. The steepness and shape of this angular gradient depends
on the particles’ type, shape and size as well as on the optical thickness of the scattering
layer. Thus, perception of circumsolar radiation by any optics pointed at the sun – be it a
pyrheliometer or a CST system – is strongly dependent not only on its opening half-angle
α but also on the sky conditions. Furthermore, the discrepancy in perception caused by
different opening angles is not constant, but also varies with sky conditions.
1Depending on the earth-sun distance, αsun varies between 0.261 and 0.271 (e.g. calculated from sunradius and earth-sun distance values given in Liou, 1980).
1.2 Aims of this Study 3
Figure 1.1: Picture of the sky, shot in the direction of the sun (courtesy of BernhardMayer). Clouds are obstructing the sun causing a distinct aureole.
1.2 Aims of this Study
Previous studies have dealt with the simulation of circumsolar radiation for a range of
atmospheric conditions (e.g. Grassl, 1971; Thomalla et al., 1983). There also have been
efforts to measure circumsolar radiation from the ground (e.g. Grether et al., 1980; Noring
et al., 1991; Ritter and Voss, 2000; Neumann et al., 2002; DeVore et al., 2009; Wilbert
et al., 2011, 2013). Following the ground measurements there have been attempts to
parameterize circumsolar radiation based on regression of meteorological variables (e.g.
Watt, 1980; Neumann et al., 1998). However these parameterizations do not consider
properly the physical reason of the variability of circumsolar radiation – the properties
of the scattering particles. The cause for circumsolar radiation, clouds and aerosols, are
temporally and spatially variable. To assess the effect of circumsolar radiation on CST
systems at new locations, data sets with good resolution and coverage in time and space
are needed to fully assess the impact on CSR system design and operations.
There have been studies which focus on the retrieval of cloud and aerosol properties from
circumsolar radiation measurements (e.g. Nakajima et al., 1983; Min and Duan, 2005; De-
Vore et al., 2009) and recently also from stellar aureole measurements (DeVore et al., 2013).
The presented study followed the opposite approach: The aim of this study was to develop
4 1. Introduction
a method which allows to derive the circumsolar radiation from globally available data on
clouds and aerosols. The methodology was developed, validated and applied on the basis
of two specific data sources: Cloud properties retrieved from measurements of the geo-
stationary Meteosat Second Generation (MSG) satellites (Sect. 2.3) and modelled aerosol
concentrations from the Integrated Forecast System (IFS) of the ECMWF2 (Sect. 3.3.1).
However, the methodology can be adapted to other data sources as well. The utilized
data sources have big advantages considering global coverage, site intra-comparability and
financial costs compared to ground measurements. The data are readily available for many
regions of the world covering longer time spans than most available ground measurement
time series of circumsolar radiation.
Considering clouds, this study only deals with circumsolar radiation caused by cirrus
clouds. These ice clouds often occur with an optical thickness τ ranging from 0.1 to
2.0, that on one hand still allows the operation of CSTs, and on the other hand makes
them detectable by satellites. Other cloud types are usually optically thicker, such that
the transmitted direct radiation is insufficient for the operation of CSTs.
The core of the method is a fast yet precise parameterization which converts cloud or
aerosol parameters into circumsolar radiation measures. The parameterization is assisted
by a lookup table which is based on simulations of the circumsolar radiation. To allow
for these simulations, the radiative transfer solver “Monte Carlo Code for the Physically
Correct Tracing of Photons in Cloudy Atmospheres”, MYSTIC (Mayer, 2009) was extended
by the capability to calculate circumsolar radiation.
Concentrating solar technologies can be divided into two categories – solar thermal units
and concentrating photovoltaics. The principle of the former is to generate heat, which can
be utilized for power generation, as industrial process heat or to drive chemical processes.
The latter uses photovoltaic cells to generate power. In solar thermal applications basi-
cally the whole spectrum of the sun light can be exploited (up to ≈ 2.5 µm, e.g. Wesselak
and Schabbach, 2009), while photovoltaics have a narrower spectral response. This work
primarily focuses on solar thermal applications, i.e. the broad band solar integrated cir-
cumsolar radiation, but the developed methods can in principle be applied to concentrating
photovoltaics with little modification.
The manuscript is structured as follows. In Chapter 2 an overview of the basic principles is
given, on which this work is founded. In Chapter 3 the tools and methods used in this study
2European Centre for Medium-Range Weather Forecasts
1.2 Aims of this Study 5
are described. This includes a description of the achieved improvements with respect to
the exact simulation of the sunshape with the radiative transfer solver MYSTIC, as well as
concerning the retrieval of cirrus cloud properties from MSG measurements with the APICS
algorithm. Finally the core of the method, the parameterization of circumsolar radiation is
developed in this chapter. Results obtained by applying the developed method on selected
data are shown in Chapter 4. Furthermore a validation utilizing ground measurements
performed at the Plataforma Solar de Almerıa, Spain, is presented. Error sources and the
respective sensitivity of the results are discussed in Chapter 5. Furthermore a comparison
of the developed method to a parameterization based on DNI and GHI measurements is
presented. A synopsis of the thesis as well as an outlook is given in Chapter 6.
Parts of this thesis concerning cirrus related circumsolar radiation have already been pub-
lished in Reinhardt et al. (2012, 2013).
6 1. Introduction
Chapter 2
Basic Principles
In this chapter the basic principles and mechanisms underlying this thesis are outlined.
Virtually all aspects of this work are connected to the transfer of sunlight through the
atmosphere. In Sect. 2.1 it is sketched how this process can be mathematically expressed
in the radiative transfer equation (RTE). Closely connected is Sect. 2.2 which deals with the
scattering of sunlight in the atmosphere. The satellite based remote sensing of properties of
cloud and aerosol particles – which are the reason for circumsolar radiation – is sketched in
Secs. 2.3 & 2.4. Sections 2.5, 2.6 & 2.7 outline how circumsolar radiation can be quantified
and how it is measured from the ground.
2.1 The Radiative Transfer Equation
The pivotal physical mechanism dealt with in this work under many different aspects is
the transfer of sunlight through the atmosphere. Let us therefore briefly recapitulate the
equations governing the radiative transfer. The following considerations are adapted from
Zdunkowski et al. (2007).
Two important radiation quantities which appear repeatedly throughout this work are the
irradiance I, also called flux, and the radiance L. The irradiance quantifies the radiant
power incident on a surface per unit area. Consequently it is given in units of (W/m2).
The radiance holds the information on how much radiant power is coming from which
8 2. Basic Principles
directions. It is given in units of (W/m2/sr). Integration of the radiance over a solid angle
Ω will yield irradiance:
I =
∫Ω
L cos θdΩ , (2.1)
where θ is the angle between the considered surface normal in regard to which I is calculated
and the solid angle element dΩ. Both quantities – irradiance and radiance – can also
be specified spectrally. Then the matching units are (W/m2/nm) and (W/m2/sr/nm),
respectively.
The problem of calculating the propagation of sunlight through the atmosphere can be
split into two parts, one for the direct radiation, i.e. the part of the radiation having not
interacted with the atmosphere, and one for the diffuse radiation. This is reasonable since
the diffuse radiation does not couple back to the direct radiation. However, the direct
radiation serves as source for the diffuse radiation.
The direct solar irradiance Idir can be calculated from the extraterrestrial flux Is according
to the Beer-Bouguer-Lambert1 law as
Idir = Is exp
− s∫0
kext(s′)ds′
(2.2)
with kext the (volume-)extinction coefficient (in units of (m−1)) and s the direct path along
which the light has traveled through the atmosphere along the direction of incidence Ω0
(comp. Zdunkowski et al., 2007, Eq. 2.32).
In this context we can also introduce the optical thickness τ which is usually defined as
the vertical integral of the extinction coefficient kext over height z. However, decisive for
the circumsolar radiation is the integral of the extinction coefficient along the line of sight
between the observer and the sun (as in Beer’s law). It is called slant path optical thickness
τs =
z2∫z1
kext(z)
µdz (2.3)
where µ is the cosine of the sun zenith angle θsun.
The diffuse solar radiation is described by an integro-differential equation (comp. Zdun-
1For the sake of brevity commonly called Beer’s law
2.1 The Radiative Transfer Equation 9
kowski et al., 2007, Eq. 2.36) which is referred to as the radiative transfer equation (RTE):
Ω · ∇Ld = −kextLd +ω0kext
4π
∫4π
P (Ω · Ω′)Ld(Ω′)dΩ′ +ω0kext
4πP (Ω0 · Ω)Idir . (2.4)
It describes the change of the diffuse radiance Ld along a direction of propagation Ω. Ω
and Ω′ are directional unit vectors. The single scattering albedo ω0 relates the extinction
coefficient to the scattering coefficient ksca and to the absorption coefficient kabs as
ω0 =ksca
kext
= 1− kabs
kext
(2.5)
and gives the relative amount of scattering on the extinction (Zdunkowski et al., 2007,
Eq. (1.49)). The scattering phase function P describes how much of the light coming from
the direction Ω′ is redirected to the direction Ω in a scattering process. Since only randomly
oriented particles are considered in this study, the bulk phase functions depend only on
the cosine of the scattering angle cos(θsca) = Ω · Ω′ (e.g. Hansen and Travis, 1974). The
first term on the right side of the RTE (Eq. 2.4) gives the reduction of radiance through
extinction, i.e. either absorption or scattering, out of the direction Ω. The second term
is a source term describing the in-scattering of diffuse radiance from other directions Ω′
into direction Ω. The third term expresses the conversion of direct radiation into diffuse
radiation through scattering.
Note that the presented equations are for monochromatic light. All optical parameters
are wavelength dependent and the equations need to be solved for every wavelength indi-
vidually (see also Sect. 3.4.3). Furthermore all variables depend on the location ~x. The
presented form of the RTE is stripped to include only processes relevant for the applica-
tions in this work. Emission of radiation which occurs in the atmosphere at considerable
amounts only in the infra-red spectrum is neglected. The transfer of radiant energy from
one wavelength to another (inelastic scattering) is also omitted, as well as it is ignored that
light can occur in different polarisation states. For a derivation of the RTE and a more
complete overview of atmospheric radiative transfer one may refer to Zdunkowski et al.
(2007).
Although the presented version of the RTE resembles already a simplified model for the
radiative transfer it still is difficult to solve. Analytically this can only be achieved if
further drastic simplifications are applied. If a more realistic version of the RTE should be
solved this must be done numerically. For this study the radiative transfer was simulated
10 2. Basic Principles
utilizing the libRadtran software package (Mayer and Kylling, 2005). In Sect. 3.4.1 the two
different RTE solvers from libRadtran which found application in this work are presented.
Unless stated otherwise the plane parallel approximation and the closely connected in-
dependent column approximation are applied in all radiative transfer simulations in this
study. The former indicates that the atmosphere is considered to consist of parallel slabs in
which the atmospheric properties do not change. That is variations are only considered in
the vertical, such that the radiative transfer collapses to a one dimensional problem. Fur-
thermore the curvature of the earth is ignored. The independent column approximation
assumes that the radiative transfer within an atmospheric column, as probed for example
by a satellite, is independent of the atmospheric properties of the neighbouring columns.
In other words, the atmosphere is treated locally as one dimensional and the radiative
transfer is calculated using the plane parallel approximation with periodic boundaries. In
general, both approximations cause errors which increase with solar zenith angle: For low
solar altitude, 3D-effects, like shadowing by clouds, are pronounced and the sphericity of
the atmosphere has to be considered (see also discussion at the end of Sect. 3.2.3).
2.2 Scattering of Sunlight
Scattering describes the change of the propagation direction of light due to interaction
with the medium. A recommendable review on scattering in planetary atmospheres was
given by Hansen and Travis (1974), but let us briefly recapitulate the most relevant aspects
concerning this study.
Forward scattering, that means a change of the propagation direction of at most a few
degrees, is the phenomenon that causes the formation of an aureole in turbid atmospheres.
Pronounced scattering in forward direction occurs on particles which are larger than the
wavelength λ of the light. For particles much larger than λ, the extinction efficiency
Qext = σext/A may become larger than one. σext is the extinction cross-section and A is
the geometrical cross-section of the particle. Accordingly Qext is the ratio of the actually
extinct radiant flux to the radiant flux hitting the geometrical cross section of a particle
(comp. Zdunkowski et al., 2007, Chap. 9.6). Qext can be as large as four but approaches
two for very large particles. This so called “extinction paradox” exists because not only
the light hitting the particle’s geometrical cross section is influenced but also the light
bypassing the particle. This is due to the wave-nature of the light and can be described
2.2 Scattering of Sunlight 11
by the diffraction theory. Therefore in the limit of large particles, the distribution of
approximately 50% of the scattered radiant energy is described by diffraction theory. For
the other half of the radiant energy, different methods, like for example geometric optics
(e.g. Petty, 2006), need to be employed.
The diffraction pattern depends on the shape of the projected area of the scattering particle.
Airy (1835) published a solution of the diffraction problem for circular apertures which is
also valid for spherical particles. In general the diffraction pattern is proportional to the
2-dimensional Fourier transform F of the projected area (Born and Wolf, 1999, chap. 8.3),
so that the diffraction phase function Pdif can be written as (DeVore et al., 2013, Eq. 3):
Pdif(θsca/λ, φ) =2π
Aλ2|F fproj(x, y)|2 , (2.6)
with θsca and φ being the scattering zenith and azimuth angle, respectively. fproj is the
pupil function and describes the particle’s projection on a surface normal to the incident
light. It is 1 inside the projection and 0 elsewhere.
In any case, diffraction predicts a primary peak of scattering into the forward direction.
This peak becomes more and more focused the larger the particles are. As can be followed
in Fig. 2.1, for a sphere the half-width-half-maximum of the forward peak is inversely
proportional to the arcus sine of the sphere’s radius r which can be approximated as 1/r
for small angles.
While the diffraction theory can be used to predict the forward scattering peak of large
particles, there is no uniform method to derive the complete scattering phase function
(θsca = 0..180) of ice crystals of arbitrary sizes and shapes (Yang et al., 2005). The
optical property datasets for cirrus clouds used in this thesis are outlined in Sect. 3.1.
They were created using an up-to-date combination of several methods.
For spherical particles, Mie theory (Mie, 1908) can be used to derive an exact solution of
the scattering properties – including diffraction effects. The optical properties for aerosol
used in this study are mainly based on Mie calculations (see Sect. 3.3).
In the limit of particles much smaller than λ (e.g. air molecules), Rayleigh theory (Lord
Rayleigh (J. W. Strutt), 1871) can be applied which predicts only a weak directional
12 2. Basic Principles
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4scattering angle /
0
10000
20000
30000
40000
50000
60000
pro
babili
ty d
ensi
ty
r =10 m
r =20 m
r =30 m
Figure 2.1: Detail of the diffraction phase functions at 550 nm for spheres with radii r of10 µm, 20 µm and 30 µm. Black lines mark the half-width-half-maximum values at 0.81,0.41 and 0.27, respectively.
dependence of the scattering. The phase function for Rayleigh scattering Pray is (e.g.
Hansen and Travis, 1974, Eq. 2.15):
Pray =3
4
(1 + cos2 θsca
). (2.7)
As Fig. 2.2 shows the phase function varies only by a factor of 2 between sideways and
forward or backward scattering. Therefore the radiation ending up in the circumsolar
region is only a small part of the scattered radiation and the contribution of Rayleigh
scattering to the circumsolar radiation can be neglected for the purposes of this study (see
also Sect. 3.5).
2.3 Remote Sensing of Clouds from MSG
2.3.1 The SEVIRI Instrument
This study relies on data of the Spinning Enhanced Visible and Infra-Red Imager (SEVIRI)
aboard the geostationary Meteosat Second Generation (MSG) satellites (Schmetz et al.,
2.3 Remote Sensing of Clouds from MSG 13
0 20 40 60 80 100 120 140 160 180scattering angle /
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
pro
babili
ty d
ensi
ty
Figure 2.2: Phase function for Rayleigh scattering.
2002) for the remote sensing of clouds. It features eleven spectral channels in the visible
and infrared spectrum with a sub-satellite-point sampling distance of 3 km and one broad
band high-resolution channel in the visible spectrum (1 km sampling distance). From the
geostationary orbit SEVIRI samples a large part of the earth which appears as disk on the
images (comp. Fig. 2.3).
The MSG-satellites are spin stabilized – that is they rotate with 100 revolutions min−1.
The main instrument SEVIRI thereby scans the earth in lines from east to west – three
lines at a time. A step motor driven mirror advances the focus of the imager by three lines
from south to north with every revolution. A full scan takes about 12 minutes. Thereafter
a calibration of the thermal channels and the repositioning of the mirror takes place so that
the instrument can sample the full disk every 15 minutes. The images in the 11 standard
channels feature 3712x3712 pixels. The time delay ∆t between the start of the scan of the
disk in the south and the probing of a certain pixel can therefore be approximated via the
pixel’s line number Nl:
∆t ≈ Nl
3712· 12 min . (2.8)
This becomes important when co-locating measurements of other instruments.
The main advantages of the instrument are the high sampling frequency combined with a
good spatial resolution which allows to follow the temporal evolution of cloud properties
anywhere within the MSG disk. The first MSG satellite (Meteosat 8) delivered operational
14 2. Basic Principles
Figure 2.3: False-color composite image of one full SEVIRI scan – the sampled part of theearth is also called “MSG-disk”.
data from 2004 on. Currently Meteosat 10 is the main operational MSG satellite and
with one further successor to be launched the MSG satellite family is expected to deliver
consistent measurements until 2018.
Especially important in the context of this work are the SEVIRI channels 1 & 3 or “VIS006”
& “NIR1.6”. They are centered at 0.635 µm and 1.64 µm and the lower and upper wave-
length boundaries are 0.56 µm and 0.71 µm, and 1.50 µm and 1.78 µm respectively. Typi-
cally the measurements in these channels are expressed as reflectivity or reflectance value
R which is basically a normalized radiance:
R =πLd2
Iλ cos(θsun), (2.9)
with L being the measured spectral band radiance, d the earth-sun distance in AU, Iλ the
spectral band solar irradiance for the channel at 1 AU and θsun the sun zenith angle at the
considered location (EUM MET TEN 12 0332).
It should be noted that SEVIRI is operated with an over-sampling strategy: While the
sampling distance is 3 km at nadir, the resolution is only about 4.8 km. As sampling is
2.3 Remote Sensing of Clouds from MSG 15
performed with constant angular stepping, this resolution decreases for off-nadir measure-
ments. Furthermore the registration of satellite pixels to latitude/longitude values may
deviate by up to 3 km at nadir with a image-to-image root-mean-square-value of 1.2 km
(Schmetz et al., 2002). These limitations must be kept in mind for example when compar-
ing SEVIRI data to ground based measurements as in Sect. 4.2.2.
A test sector of 189x252 satellite pixels within the whole MSG disk was defined (see
Fig. 2.4). The evaluation of SEVIRI measurements was limited to this sector for the sake
of reduced computing time and storage space. The sector includes the southern part of
the Iberian Peninsula where recently the construction of concentrating solar power plants
flourished. Currently approximately 40 solar power plants are in operation in this area.
Also the CST research center Plataforma Solar de Almerıa (PSA) is inside the sector, as
well as parts of northern Africa, where an increasing amount of CST projects are planned.
The chosen area includes Mediterranean as well as desert climate zones and should there-
fore offer suitable test conditions. Ocean pixels and pixels containing continental water
surfaces were excluded from evaluation using a land/water mask from the EUMETSAT
Land Surface Analysis Satellite Applications Facility (LANDSAF).
2.3.2 Retrieval of Cloud Properties from Reflected Solar Radia-
tion
The Algorithm for the Physical Investigation of Clouds with SEVIRI (APICS) is used
in this study to derive cloud properties from the SEVIRI measurements. APICS is an
implementation of the method developed by Nakajima and King (1990) for SEVIRI. The
method uses measurements of reflected solar radiation to retrieve the optical thickness and
the effective radius reff of clouds.
The effective radius is a parameter that is commonly used to characterize the scattering
and absorption properties of disperse particle size distributions by means of a single scalar
(Hansen and Travis, 1974; McFarquhar and Heymsfield, 1998). For spherical particles
like water droplets it is defined as the ratio of the third to the second moment of the
size distribution. There is however an ambiguity in the definition of reff for non-spherical
particles. In this study it is defined as
reff =3∫V (L)n(L)dL
4∫A(L)n(L)dL
(2.10)
16 2. Basic Principles
Figure 2.4: Map covering the test sector. SEVIRI pixels are evaluated inside the sectorover land only. The cross marks the location of the the research center Plataforma Solarde Almerıa.
2.4 Passive Satellite Remote Sensing of Aerosol 17
where n is the number concentration, L the maximum dimension, V the volume and A
the projected area of the particles. This definition was also used in the generation of the
utilized cirrus bulk optical properties (e.g. Baum et al., 2005a) which are described in
Sect. 3.1.
The method by Nakajima and King (1990) exploits the spectral course of the absorption
coefficient of water and ice. While cloud particles in the visible part of the spectrum
basically solely scatter sun light, they also absorb light in the infrared. The mean path
length traveled by light through a cloud particle during a scattering event depends on
the particle size. Alike, the absorption depends on the particle size. This is reflected in
the single scattering albedo ω0 which is basically 1 in the visible part of the spectrum
independent of particle size but < 1 with a dependence on reff in the infrared (comp.
Fig. 2.5). Thus, a measurement of reflected sun light in the visible is only sensitive to
the amount of scattering by the cloud particles and therefore the optical thickness. A
measurement in the infrared is sensitive to both, scattering and absorption. It therefore
contains information about the optical thickness and the particle size. By performing a
measurement in both parts of the spectrum, one can unravel both informations. For now
we leave it at reading from Fig. 2.5 that the SEVIRI channels2 1 & 3 can be used to perform
such a retrieval. The technical details of the APICS retrieval are outlined further down in
Sect. 3.2.
2.4 Passive Satellite Remote Sensing of Aerosol
Aerosols scatter and absorb sunlight. The share that is backscattered into space can be
used to remotely sense aerosol properties. However aerosol can be comprised of a large
variety of particles in terms of sizes, material and shape. This diversity alone makes remote
sensing challenging.
The signal reaching the satellite is affected not only by the aerosol but also by the at-
mosphere and often to large amounts by the surface. A rigorous cloud screening must be
applied to the satellite measurements, else the aerosol amount will be overestimated. This
is problematic since even sub-scale clouds or very thin clouds which are hard to detect will
affect the retrieved aerosol properties. The surface reflectivity must be known precisely
since it accounts for a large share of the reflected sun light. But even with a well charac-
2Filter functions in Fig. 2.5 from EUMETSAT document EUM/MSG/TEN/06/0010
18 2. Basic Principles
0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0wavelength / m
0.6
0.7
0.8
0.9
1.0
single
sca
tteri
ng a
lbedo
reff =5 m
reff =15 m
reff =30 m
reff =60 m
reff =90 m
0.0
0.2
0.4
0.6
0.8
1.0
SEV
IRI se
nsi
tvit
y
Figure 2.5: Spectral course of the single scattering albedo ω0 for solid-column particlemixtures for different effective radii reff . Shown in grey are the filter functions for thechannels 1 & 3 of the SEVIRI instrument aboard Meteosat-8.
terized surface reflectivity it can be hard to retrieve aerosol properties. This is the case
if the aerosol has similar scattering and absorption characteristics as the surface: While
over a dark surface like the ocean aerosol will increase the satellite-measured reflectivity, it
may – depending on the aerosol absorption – decrease the reflectivity over bright surfaces
like snow or ice. In between, the sensitivity of the satellite-measured reflectivity on the
atmospheric aerosol load becomes zero. This is often the case for desert regions: Uplifted
dust over a desert can be hardly quantified.
Todays retrievals which go beyond the determination of the optical thickness alone use a
synergy of measurements, that is multi-angle, multi-spectral and polarized measurements.
This way some information about the aerosol type and size distribution can be obtained.
However even these sophisticated retrievals do not provide full global coverage. For example
the SYNAER retrieval, which is a multi-spectral retrieval for the aerosol optical thickness
and aerosol type, does not work over deserts and the aerosol type is only given if the aerosol
optical thickness is larger than 0.1 (Holzer-Popp et al., 2008). Furthermore multi-angle or
polarized measurements are performed only from low earth orbit satellites such that the
temporal resolution of todays sophisticated aerosol retrievals is typically in the order of
days.
To obtain a data set without gaps, that also contains data for arid regions which are of
2.5 The Sunshape 19
special interest for CST, I decided to use modeled aerosol data from the Integrated Forecast
System (IFS) of the European Centre for Medium-Range Weather Forecasts (ECMWF).
The details concerning the model are outlined in Sect. 3.3.1.
2.5 The Sunshape
While a complete characterization of the instantaneous circumsolar radiation can only be
given by a 2D radiance distribution, it is commonly reduced to a one-dimensional radial
radiance profile assuming radial symmetry. Such a radial radiance profile from the center of
the sun outward is called sunshape. It is an adequate simplification if scattering is caused
within atmospheric layers that are horizontally homogeneous within the field of view. At
large sun zenith angles θsun, a 1D profile might not be appropriate, even for homogeneous
conditions, as significant differences in the path length through the scattering layer occur,
even within a small field of view.
The projection of a circular field of view onto a horizontal (scattering) layer is an ellipse. It
has two length scales L1 and L2 corresponding to its major and minor axis. Both depend
on the elevation of the scattering layer H and on the sun zenith angle θsun:
L1 = H/ tan(90 − θsun − αcir)−H/ tan(90 − θsun + αcir) (2.11)
L2 = 2 sin(αcir)H/ cos(θsun) . (2.12)
Figure 2.6 shows that, even if only a field of view with αcir = 2.5 is considered, the
length scales can be in the order of several kilometers for cirrus clouds at an elevation of
H = 12 km. Cirrus clouds normally show variability on smaller scales so that a sunshape is
in general not a good representation of the instantaneous 2D radiance distribution around
the sun. It is therefore advisable to temporally average sunshapes measured from the
ground since advection of the clouds through the field of view will normally smooth the
average radiance distribution.
For aerosols the sunshape is a better representation of the instantaneous 2D radiance
distribution for two reasons. First, aerosols normally show considerably less horizontal
variability than cirrus clouds. And second, most of the aerosol is constricted to the bound-
ary layer which has only a few kilometers vertical extent at most, so that often the length
20 2. Basic Principles
scale of the field of view is a magnitude smaller than for cirrus clouds.
Figure 2.6: Length of the major and minor axis of the ellipse that forms as a circular fieldof view with αcir = 2.5 intersects a horizontal layer at 12 km elevation under different sunzenith angles.
Figure 2.7 shows exemplary three sunshapes simulated with the radiative transfer solver
MYSTIC (see Sect. 3.4.2). In case of a clear atmosphere the radiance inside the sun
disk features the highest values compared to the other simulations, while it exhibits the
lowest values in the circumsolar region. After adding an aerosol layer to the simulation
the radiance inside the sun disk decreases by ≈ 40%, while in the circumsolar region it
increases by approximately two magnitudes. If the aerosol is replaced by a cirrus layer of
the same optical thickness, the sunshape looks quite different in the circumsolar region:
Close to the sun the radiance is once more increased by about two magnitudes but falls
of much more steeply with increasing angular distance to the sun. This difference in the
sunshapes reflects the different particle sizes: The ice particles are larger than the aerosol
particles, therefore the forward peak in their scattering phase function is more focused.
To give the reader an impression of corresponding irradiance values, the sunshapes were
integrated over three angular intervals; once over the sun disk (0.0 – 0.266), once over
the circumsolar region up to 1.0 from the sun center and once over the annular region
from 1.0 to 2.5 from the sun center. While 1.0 is meant to represent the field of view
of a CST receiver, 2.5 is the opening half-angle standard of pyrheliometers by the World
2.6 The Circumsolar Ratio – CSR 21
Meteorological Organisation (WMO) (CIMO Guide). Table 2.1 lists the results of the
integration in W/m2. The cirrus with the pronounced forward scattering peak produces
higher irradiance values than the aerosol layer, especially within the sun disk and the
circumsolar region closely around the sun. Compared to aerosol or clouds, air molecules
only cause negligible irradiance in the circumsolar region.
Considering once more the sunshapes in Fig. 2.7 one can realize that the brightness in the
sun disk itself (α < 0.266) is not distributed evenly. The center is brighter than the rim.
This effect is called limb darkening. It is not only caused by the terrestrial but rather by the
solar atmosphere: Most of the sun light reaching the earth is emitted in the photosphere
of the sun, which besides emission also shows strong absorption of light. As the sun is a
sphere, light reaching the earth from the “rim” travels along a slanted path through the
photosphere, while light from the disk center exits the photosphere perpendicular. The
mean depth in the photosphere, from which light reaching the earth originates, decreases
therefore from the disk center to the rim. The temperature and therefore also the emission
of light in the photosphere increases with depth. Therefore the sun disk appears brighter
in the center than at the rim (Scheffler and Elsasser, 1990). The implementation of the
limb darkening effect in the radiative transfer solver MYSTIC is outlined in Sect. 3.4.2.
Table 2.1: Shortwave integrated irradiance values for the three sunshape simulations shownin Fig. 2.7. Values are given for the sun disk (0.0 – 0.266), the circumsolar regionextending to 1.0 angular distance from the sun center and an annular region extendingfrom 1.0 to 2.5 angular distance from the sun center.
0.0 – 0.266 0.266 – 1.0 1.0 – 2.5
Rayleigh atm. 998 W/m2 0.01 W/m2 0.06 W/m2
Dust aerosol 621 W/m2 3.2 W/m2 11 W/m2
Cirrus 684 W/m2 90 W/m2 20 W/m2
2.6 The Circumsolar Ratio – CSR
The characterization of the circumsolar radiation can be simplified further by introducing
a scalar quantity – the circumsolar ratio (CSR). It is defined as the normal irradiance
coming from an annular region around the sun divided by the normal irradiance from
this circumsolar region and the sun disk. For the remainder of the document the term
irradiance always refers to normal irradiance, i.e. irradiance on a surface perpendicular
22 2. Basic Principles
0.0 0.5 1.0 1.5 2.0 2.5 3.0angular distance from sun center α / °
101
102
103
104
105
106
107
bro
ad b
and r
adia
nce
/ W
/m2/s
r
0.0 0.5 1.0 1.5 2.0 2.5 3.0angular distance from sun center α / °
101
102
103
104
105
106
107 Pure Rayleigh Atm.
OPAC desert dust OD=0.5
Ice, solid col., Reff=40mu, OD=0.5
Figure 2.7: Left panel: MYSTIC simulation of sunshapes for three atmospheric situations(MYSTIC, see Sect. 3.4.2). A pure Rayleigh atmosphere (only molecules, blue), with alayer of OPAC desert dust aerosol (see Sect. 3.3.2) with an optical thickness at 550 nm of 0.5(green) or with a cirrus composed of solid-columns (see Sect. 3.1) with reff = 40 µm (red).Right panel: Same as left but the circular/annular regions marked for which irradiancevalues are given in Tab. 2.1.
to the direction pointing at the sun. The disk angle αsun gives the extent of the sun disk
measured from its center to the edge. The circumsolar region reaches from the sun’s edge
to the angle αcir (> αsun) which is subject to arbitrary definition and measured from the
sun’s center as well. With this we can write
CSR(αcir) =
2π∫0
αcir∫αsun
L(α) cos(α) sin(α)dαdφ
2π∫0
αcir∫0
L(α) cos(α) sin(α)dαdφ
(2.13)
where L is the sunshape – i.e. the radiance depending on the angular distance α from
the sun center. The cosine term is safely neglected in the following since even for the
maximum αcir of 5 considered in this study the errors due to this are smaller than 0.2%.
This estimate was derived by evaluating the error for a worst case sunshape which does
2.7 Ground Based Measurements of the Circumsolar Radiation 23
not decrease with increasing distance to the sun but stays constant (L = const). The error
is in this case5∫0
(1− cosα) sinα dα
5∫0
cosα sinα dα
= 0.2% . (2.14)
Buie et al. (2003) approximated the sunshape in the circumsolar region as a power-law
function which can be considered a line in a log-log diagram. They developed a fit for
the two parameters intercept and gradient of this function in terms of the CSR, based
on broad band solar integrated field measurements of the sunshape. Within the sun disk
(α ≤ 4.65 mrad = 0.266) they used a fixed shape. In total they parameterized the
normalized sunshape Φ as
Φ(α) =L(α)
L(α = 0)=
exp(κ)αγ, for α > 4.65 mradcos(0.326α)cos(0.308α)
, for α ≤ 4.65 mrad(2.15)
with
γ = 2.2 ln(0.52 · CSR) · CSR0.43 − 0.1 , (2.16)
and
κ = 0.9 ln(13.5 · CSR) · CSR−0.3 . (2.17)
The angular distance to the sun center α is expressed in mrad. Employing this fit, the
CSR can be used to characterize the sunshape, which is however not utilized in this study.
2.7 Ground Based Measurements of the Circumsolar
Radiation
Circumsolar radiation in terms of the sunshape is intricate to measure because a high
dynamic range, a good angular resolution and a precise sun tracking are required at the
same time. A cornerstone of ground based circumsolar radiation measurement is the so
called “LBL reduced data base” which contains about 180 000 sunshapes from 11 U.S.
sites measured with four identical devices in the late 1970s and early 1980s (Noring et al.,
1991). However LBL’s measurement system was a scanning telescope resulting in a rather
low angular resolution of about 0.025. More recently CCD camera based systems have
24 2. Basic Principles
been developed allowing to capture the 2D radiance distribution at once in high angular
resolution (e.g. Neumann et al., 1998; Neumann and Witzke, 1999; Ritter and Voss, 2000).
Data from a system using a CCD camera is also used in this study to validate the results.
This system is based on the sun and aureole measurement system (SAM) which was orig-
inally developed for the retrieval of cloud properties (DeVore et al., 2009) and is available
commercially from Visidyne Inc.3. Apart from a SAM the system constitutes of a Cimel
CE-318N EBS9 sun photometer and a dedicated post-processing software. The sun pho-
tometer is also part of the “Aerosol Robotic Network – AERONET” (Holben et al., 1998)
and products from the AERONET processing scheme are also used later in the validation
section (Sect. 4.2.1). SAM measures the sunshape at a wavelength selected with a filter.
The post-processing software uses the additional information gathered by the sun photome-
ter to convert this spectral sunshape to other wavelengths or to broad band solar integrated
values. Furthermore, this software also calculates CSR values from the obtained radiance
profiles. Sunshapes simulated with the adapted MYSTIC version (Sect. 3.4.2) have been
used to validate the post-processing. The system and its validation is described in detail
in Wilbert et al. (2013). For the sake of brevity it will be called SAMS (for SAM based
System) in the following. While in principle three SAMSs exist (one at the Plataforma So-
lar de Almerıa, Tabernas, Spain, one in Odeillio, France, and one in Masdar, Abu Dhabi),
only the data of the system in Tabernas were used (location marked in Fig. 2.4). The
quality assurance of the SAMS data is an elaborate process and only the intensive contact
with Stefan Wilbert – the principal investigator of the Tabernas SAMS – allowed to detect
and remove some erroneous values in the original dataset. SAMS data from the years 2011
and 2012 have been used in this study. Before 04 Jul. 2011 SAMS was operated with a
670 nm filter which was then replaced by a 870 nm filter. The SAMS dataset provided by
Stefan Wilbert also includes DNI and GHI from the collocated meteorological station at
the PSA.
Pursuing the goal of making circumsolar radiation measurements cheaper and less mainte-
nance dependent, some studies have been presented recently which show the potential of
new circumsolar radiation instruments: Wilbert et al. (2012) investigated a pyrheliometer-
like instrument which allows to measure with different sized apertures. Kalapatapu et al.
(2012) presented a sunshape profiling irradiometer which is based on the principle of the
rotating shadowband radiometer.
3http://www.visidyne.com
Chapter 3
Tools and Methods
This chapter deals with the utilized and developed tools and methods. First, the optical
property datasets for cirrus clouds which found application are described in Sect. 3.1. In
the following Sect. 3.2 the APICS cloud property retrieval framework is introduced and
achieved improvements are outlined. Next, the aerosol input from ECMWF’s IFS and its
processing is specified in Sect. 3.3. Section 3.4 motivates the use of the radiative transfer
solver MYSTIC in this study and outlines how it was adapted. Section 3.5 is of special
importance, as the method to parameterize circumsolar radiation is developed therein.
Finally, Secs. 3.6 & 3.7 provide sensitivity studies concerning the shape of the scattering
particles and the hygroscopic growth of aerosol, respectively.
3.1 Optical and Microphysical Properties of Cirrus
Clouds
The ice crystals within cirrus clouds feature a variety of shapes. The particle shape influ-
ences the optical properties of the ice crystals like the scattering phase function and the
single scattering albedo. This in turn influences the modelling of the circumsolar radiation
as well as the cloud property retrieval which in the end is based on radiative transfer mod-
elling as well. Passive satellite instruments with only one viewing direction do not provide
information about the ice particle shape composition within a cloud. Therefore it has to
be assumed a priori.
To be able to assess the uncertainty that is caused by this assumption, a range of different
26 3. Tools and Methods
cloud bulk optical properties were used in this study for the modelling of the circumsolar
radiation as well as for the cloud property retrieval. Optical properties for clouds featuring
a realistic particle shape mixture were considered as well as for clouds composed of particles
of only one single shape. While the latter are not necessarily a realistic assumption they
are meant to represent the extremes in the natural variability of the cloud’s composition.
The particle mixtures and associated optical properties are described in Baum et al.
(2005a), Baum et al. (2005b) and Baum et al. (2011). In the following the older version
is referred to as “Baum v2” and the newer version as “Baum v3.5”. The latter mixture
incorporates more particle shapes and the particle surface is “severely roughened” while
“Baum v2” is composed of particles with smooth surfaces. The five different single particle
shapes that are considered in this study comprise solid- and hollow-columns, aggregates
composed of eight hexagonal columns, planar bullet rosettes, and droxtals. They all have
smooth surfaces like in “Baum v2”. Hong Gang created single scattering properties using
the improved geometric optics method (Yang et al., 2000) for these particles – valid for nar-
row size bins. From these Claudia Emde computed bulk optical properties by integrating
the binned single scattering properties over gamma size distributions
n(D) = ND1b−3 exp
(−Dab
), (3.1)
with the parameters a and b. N is a normalisation constant and the measure of particle
size D is defined as
D(L) =3V (L)
2A(L), (3.2)
where L is the maximum dimension, V the volume and A the projected area of the particles.
The parameter b was fixed to 0.25 while a was iteratively determined such that the size
distribution yields the desired effective radius (Eq. 2.10) (pers. comm. Claudia Emde,
2012, Bugliaro et al., 2013). Correspondent to the contributors Hong, Emde, Yang the
optical properties for the individual particle shapes are referred to under the acronym
HEY. All optical property datasets cover an effective radius range of 5 µm – 90 µm, except
for “Baum v3.5” which only extends from 5 µm – 60 µm.
3.2 Cloud Property Remote Sensing with the APICS Algorithm 27
3.2 Cloud Property Remote Sensing with the APICS
Algorithm
The following section gives an overview of the APICS cloud property retrieval and outlines
how it was modified during this study to improve the retrieval in regard to optically thin
cirrus. The section is partly adapted from Reinhardt et al. (2013).
3.2.1 Algorithm Description
This study relies on the “Algorithm for the Physical Investigation of Clouds with SEVIRI“
(APICS) framework to derive effective radius and optical thickness of cirrus clouds. APICS
(Bugliaro et al., 2011, 2012, 2013) implements a cloud property retrieval based on Nakajima
and King (1990) (basic principle see Sect. 2.3.2) for the SEVIRI channels 1 and 3 in the
solar spectrum (centered at 0.6 µm and 1.6 µm).
The cloud property retrieval uses lookup tables (LUTs) in which pre-simulated reflectivity
values for the two SEVIRI channels (R006 & R016) are stored as a function of the most
relevant parameters, namely as a function of the sun and satellite zenith angles, the relative
azimuth between sun and satellite, the albedo in both channels, the particle size in terms
of the effective radius and the optical thickness. For each ice particle shape (mixture) an
individual LUT is used. Simplifications and assumptions about the state of the atmosphere
are required, because the two independent pieces of information (the two satellite channels)
allow the retrieval of only two quantities, in our case optical thickness and effective radius.
In principle the retrieval returns the (τ, reff)-combination which minimizes the residual be-
tween the pre-calculated and measured reflectivity values, which is calculated as (Bugliaro
et al., 2013)
χ2 :=
(R006meas
R006lut
− 1
)2
+
(R016meas
R016lut
− 1
)2
. (3.3)
The user can choose between different algorithms to minimize χ2. For this study APICS
was operated with an “Powell’s Dog-Leg”-algorithm which was adapted from Madsen et al.
(2004) (pers. comm. Luca Bugliaro, 2013). It is a fast iterative algorithm to find a local
minimum. The LUTs are interpolated between the supporting points; for this study the
retrieval was setup to use a spline interpolation.
For the operation of APICS in this study it was assumed that the channels 1 and 3 of the
28 3. Tools and Methods
SEVIRI instrument aboard the operative “Meteosat 9”-MSG satellite exhibit an underesti-
mation of about 6% (Ham and Sohn, 2010) and an overestimation of 2% respectively (Pers.
comm. Philip Watts, EUMETSAT, 2009). Therefore, a pre-processing step was applied to
take care of the corresponding recalibration of reflectance values. Recently, Meirink et al.
(2013) published similar calibration coefficients. In a satellite inter-calibration study they
found an underestimation of 8% and an overestimation of 3.5% for the SEVIRI channels 1
and 3, respectively.
3.2.2 Auxiliary Albedo Dataset
While APICS was originally developed as a multi-purpose cloud property retrieval, it
is especially important in the context of this study that optically thin cirrus clouds are
retrieved at the best. To account for that, APICS was modified in a common effort of Luca
Bugliaro and myself as described in the following.
The most important sources of uncertainty, considering thin ice clouds, are the a priori
assumptions about the ground reflectivity, the aerosol properties and the ice particle shape.
Figure 3.1 illustrates that a wrong assumption about the ground reflectivity can lead to
large errors in the retrieved cloud properties, especially for clouds with τ < 5. Both panels
in the figure show excerpts of the lookup tables for albedo value combinations which were
actually observed within the considered test sector (Fig. 2.4). The albedo combination in
the left panel, for example, was observed for a densely forested region in Spain, just north
of Gibraltar (Alcornocales national park), while the albedo combination in the right panel
is typically observed for the desert regions in the south-east of the test sector. We adapted
APICS such that initially existing problems concerning wrong assumption of the ground
reflectivity are reduced.
APICS relies on the albedo to describe the ground reflectivity – that is, isotropic ground
reflection is assumed. The albedo must be determined a priori for both SEVIRI channels at
every pixel. The original version of APICS used albedo products based on measurements
of the Moderate Resolution Imaging Spectroradiometer (MODIS) aboard polar orbiting
satellites – either the “blacksky” or “whitesky” albedo from the Ambrals processing scheme
(Strahler et al., 1999). Now we modified APICS so that it generates its own self-consistent
albedo product. This product is based on the SEVIRI “Clear Sky Reflectance Map”, which
is the average of the reflectance for a given time of day over the seven preceding days un-
der clear sky conditions (EUM OPS DOC 09 5165). The clear sky reflectance contains
3.2 Cloud Property Remote Sensing with the APICS Algorithm 29
0.0 0.2 0.4 0.6 0.8 1.00.6 µm Reflectivity
0.0
0.1
0.2
0.3
0.4
0.5
0.6
1.6
µm
Refl
ecti
vit
y
5 µm
10 µm
15 µm
20 µm
25 µm
30 µm
35 µm
40 µm
50 µm
60 µm
90 µm
0.0
0.8
1.6
3.0
5.010.0 20.0 40.0 100.0
0.0 0.2 0.4 0.6 0.8 1.00.6 µm Reflectivity
0.0
0.1
0.2
0.3
0.4
0.5
0.6
1.6
µm
Refl
ecti
vit
y
5 µm
10 µm
15 µm
20 µm
25 µm
30 µm
35 µm
40 µm
50 µm
60 µm
90 µm
0.0
0.8
1.6
3.0
5.0
10.0 20.0 40.0 100.0
Figure 3.1: Details of the APICS lookup table for “Baum v2” for a sun zenith angle of30, a relative azimuth between sun and satellite of 20 and a satellite zenith angle of 41.4.Blue: Lines of equal effective radius. Dashed red: Lines of equal cloud optical thickness τ .Left panel: Albedo combination for the 0.6 µm and 1.6 µm-channels of (0.10, 0.25). Rightpanel: Albedo combination of (0.35, 0.60). Plotting script courtesy of Luca Bugliaro.
contributions from the ground as well as from the atmosphere. From this combined signal
the ground albedo needs to be extracted. The albedo values in both SEVIRI channels are
therefore retrieved using the APICS lookup tables. To this end it is evaluated for which
albedo the pre-simulated reflectivity values match the clear sky reflectance best, where
τ = 0. This procedure has the advantage of being consistent with the cloud property
retrieval concerning atmospheric gas composition, radiative transfer modelling and instru-
ment calibration. Furthermore we assume that the thus retrieved albedo also corrects for
some of the long term variability in the aerosol properties which may deviate from the
constant aerosol properties used in the simulations for the APICS lookup tables. Long
term variability here denotes changes on the scale of a few days to one week and which
are therefore slow enough to influence the clear sky reflectance averaged over seven days.
The “Clear Sky Reflectance Map” is routinely only available for 12 UTC. Therefore the
albedo is only retrieved at this time and used for the whole day. Nevertheless the success
rates (defined in the following) reached by APICS with this method are superior compared
to using the MODIS albedo datasets. In principle one could develop a product alike the
30 3. Tools and Methods
“Clear Sky Reflectance Map” with a day-of-time dependence, but this was assessed to be
beyond the scope of this study.
Because of the uncertainties in the a priori assumptions described above and because of
other insufficient projections of the reality (e.g. 1D radiative transfer calculations) the cloud
property retrieval results can be erroneous. Often the measured reflectivity pair cannot
even be reproduced by any parameter combination in the lookup tables. In these cases the
retrieval will return a (τ, reff) combination from the “edge” of the LUT and χ2 – although
minimized – will differ from zero. Let us call these occurrences “outliers”. Please note that
the definition of outliers is very strict. In principle one could add all those data points
to the “hits”, which lie within the uncertainty range of the retrieval due to errors in the
measurements and in the assumptions. These errors are, however, difficult to quantify.
Since the number of outliers is used in the following only as a relative quality criterion,
the exact definition is not relevant. The success rate of the cloud property retrieval S is
defined as one minus the ratio of the number of outliers to the total number of considered
pixels in a scene.
S = 1− outliers
considered pixels(3.4)
The success rate was calculated for four months of SEVIRI measurements within the
test sector (see Fig. 2.4), which should be characteristical for the four different seasons
(Jun/2011, Sep/2011, Dec/2011 and Mar/2012). The average success rates are 62%, 62%
and 81% for the MODIS “blacksky”, MODIS “whitesky” and the APICS-generated albedo
respectively when using the “Baum v2” optical properties. Considering the better values
with the APICS-generated albedo, APICS was operated with this new albedo product
for the retrieval of circumsolar radiation. Note that success means the retrieval obtained
values, which need not be correct, though. It was decided to include the “outliers” in
the circumsolar radiation calculations with the (τ, reff) combination that APICS returns
because they are the best possible result for a given set of measurements.
As one can see from the not perfect success rate this approach attenuates the problems with
wrong a priori assumptions but does not solve them completely. The APICS-generated
albedo product can adapt to changes of the real surface albedo only on the scale of a
few days while the real changes – e.g. due to precipitation – can be on the scale of hours.
Further reading on the influence of inhomogeneous surface albedo on Nakajima & King-like
cloud property retrievals can be found in Fricke et al. (2012).
3.2 Cloud Property Remote Sensing with the APICS Algorithm 31
3.2.3 Cirrus Cloud Mask
The APICS retrieval is only performed for pixels classified as cirrus. This is mandatory
since else a measurement of increased reflectivity compared to the pre-computed clear
sky value at 0.6 µm will be misinterpreted by the algorithm as a cirrus cloud, even if
the considered pixel is completely cirrus free. Such increased reflectivity values can, for
example, be caused by high aerosol loadings or water clouds.
APICS originally relied on a cirrus cloud mask generated by the “Meteosat Second Gener-
ation Cirrus Detection Algorithm” v2 (MeCiDa) (Krebs et al., 2007; Ewald et al., 2013).
However, it was foreseen that the retrieval of circumsolar radiation would benefit if the
cirrus cloud mask used in conjunction with APICS was improved considering the detection
efficiency of optically thin clouds. During the course of this study another cirrus cloud
property retrieval algorithm was developed at DLR, which is called “Cloud Optical prop-
erties derived from CALIOP and SEVIRI” (COCS) (Kox et al., 2011; Kox, 2012). It is
very sensitive to optically thin cirrus clouds and was finally used to replace the original
cirrus cloud mask algorithm.
COCS uses a neural network approach to convert the measurements in the infrared channels
of SEVIRI into the two parameters ice optical thickness and cloud top pressure. The neural
network was trained with a collocated dataset of SEVIRI observations and retrieval results
for the “Cloud-Aerosol Lidar with Orthogonal Polarization” (CALIOP) aboard the polar
orbiting satellite CALIPSO1. CALIOP is very sensitive to cirrus clouds and therefore the
neural network delivers especially good results for thin clouds with τ < 1. However the
effective radius, which has strong influence on the circumsolar radiation (see Sect. 3.6), is
not retrieved. Furthermore, in an intercomparison of retrieved optical thickness between
COCS and APICS, COCS showed a distinct bias to lower optical thickness for cirrus clouds
for which APICS retrieved τ > 1. Because COCS is known to saturate at τ ≈ 2.5 (Kox,
2012), I assume that the optical thickness from APICS is more reliable for clouds with
τ > 1. For these reasons, the output from COCS was only used to generate a cloud mask
for APICS.
Ostler (2011) and Bugliaro et al. (2012, 2013) computed the detection efficiency ηdet of
MeCiDa utilizing airborne LIDAR observations. MeCiDa detects virtually all of the cirrus
clouds with τ > 0.5 but only about 50% at τ ≈ 0.2. For most CST applications only clouds
1Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations
32 3. Tools and Methods
with a slant path optical thickness τs = τ/ cos(θsun) smaller than 3.0 are relevant (with θsun
being the sun zenith angle), otherwise too much light is extinguished in order to allow for
energy production. Therefore, clouds with τ < 0.5 account for a good share of the relevant
clouds. The same studies show that COCS has advantages in detecting these clouds and
reaches a detection efficiency of 80% even at τ = 0.15 while the detection efficiency of
MeCiDa falls below 80% already between τ = 0.3 and τ < 0.4. It was therefore decided to
derive a cloud mask from COCS results. All pixels that are assigned an optical thickness
larger than 0.1 by COCS are assumed to be cloudy. This cut-off criterion of τ > 0.1 is
necessary to keep the false alarm rate (FAR) at an acceptable level. Kox (2012) assessed
the false alarm rate to be 5% for this cut-off using CALIOP observations as reference. The
impact of the change from MeCiDa to COCS on derived circumsolar radiation is discussed
in Sect. 5.2.1.
It should be kept in mind that there is no universally valid method to generate a cirrus
cloud mask. Different methods differ in detection efficiency and false alarm rate. Often a
higher detection efficiency goes along with increased false alarm rate. Furthermore these
quality measures are subject to almost arbitrary definition: Consider only the treatment
of satellite pixels that contain water and ice clouds at the same time. Should they be
classified as cirrus cloud or not? This definition alone will strongly influence ηdet and the
FAR of any algorithm. The approach in this study was to detect as many cirrus clouds
as possible and at the same time to minimize the impact of false detections – which is
possible only to a certain extent: The re-calibration of the SEVIRI measurements and the
new albedo product help to minimize misinterpretation in clear sky conditions. However,
they do not address the problems with water clouds (see also Sect. 4.2.2).
Because COCS uses SEVIRI’s infrared channels, it is independent of the sun’s position.
The APICS cloud property retrieval however relies on reflected sun light, which causes
deteriorations at sunrise and sunset as the plane parallel and the independent column
approximation break down. Loeb et al. (1997) showed that for broken clouds with high
vertical extent or for stratiform clouds with bumpy cloud tops the plane parallel approx-
imation causes considerable errors even for sun zenith angles as small as 65. However,
in a previous study Loeb and Davies (1997) concluded that the errors are in general less
pronounced for optically thin clouds. In this study APICS results are used for sun zenith
angles up to 80. This can be justified as the considered cirrus clouds are optically thin-
ner and, as I assume, normally more homogeneous than the strato-cumulus like synthetic
clouds investigated by Loeb et al. (1997).
3.3 Aerosol Input and Processing 33
3.3 Aerosol Input and Processing
3.3.1 Modeled Aerosol Data from the ECMWF IFS
The aerosol data from which circumsolar radiation was derived in this study stem from
the Integrated Forecast System (IFS) of the European Centre for Medium-Range Weather
Forecasts (ECMWF). It is a global weather model which incorporates prognostic aerosol
variables. These variables are also considered in the data assimilation process used to
initialize the model: Besides satellite-measured reflectivity values at aerosol-relevant wave-
lengths, the MODIS aerosol optical depth product at 550 nm (Collection 5) is assimilated.
The IFS was chosen for several reasons: It provides global coverage; it has large support
of the scientific community and several major scale projects rely on it (GEMS, MACC,
MACC II); furthermore it serves as driver for several local air quality models who use the
IFS output as boundary condition. Therefore there has been quite some effort to validate
the model (Mangold et al., 2011; Bouarar et al., 2012; Benedictow et al., 2012). The most
important point is the assimilation of remote sensed aerosol data: Thanks to it, the MODIS
aerosol optical thickness (AOT) should be well represented in the model results. Where
however no information from satellite is available, the model parameterizations and physics
step in to fill the gap. Mangold et al. (2011) conclude that the assimilation of the MODIS
AOT significantly improves the model performance especially concerning AOT-peaks and
in general even over areas without MODIS data (bright desert surfaces). This is especially
important since many regions of interest for CST applications have arid/desert climate.
The IFS does not predict the aerosol particle size explicitly, which is a crucial parameter
for the circumsolar radiation. However, the concentrations of several aerosol types with
(fixed) size distributions are calculated – where appropriate even subdivided into several
size bins (Sect. 3.3.2). The linear combination of these aerosol types thus implicitly holds
the information about the aerosol particle size.
Aerosol is represented as different types in the IFS (dust, sea salt, organic matter, black
carbon and sulphate). The generation of dust and sea salt aerosol is parameterized from
model variables which are basically the 10 m-wind speed plus surface characterizing vari-
ables for dust, for example the soil moisture. The other aerosol sources are independent of
the meteorology and stem from emission inventories. For the removal, dry and wet deposi-
tion is considered consistent to the meteorological variables. The aerosol types containing
34 3. Tools and Methods
large particles, namely dust and sea salt, are represented via three size bins so that a coarse
differentiation in settling speed can be achieved. Furthermore the hygroscopic shares of the
aerosols are allowed to grow with increasing relative humidity. This short description of
the aerosol modelling within the IFS sums up only the most relevant facts of the detailed
description from Morcrette et al. (2009).
The data used in this study for the validation of the method (Sect. 4.2.1) stem from a so
called “near real time” (NRT) forecast performed with the IFS. It has the drawback that
the complete set of model variables is not is available as output. The information on relative
humidity is not included. However, the re-analysis dataset, which would be preferable and
which was intended to be used in deriving global characteristics of circumsolar radiation,
was not available for the time spans for which ground based measurements were obtainable.
Because the validation showed dissatisfactory results when processing the model output
head on, the re-analysis dataset found no application in the end (see Sects. 4.2.1 & 5.1).
Details concerning the model setup can be found in (Stein et al., 2011). The NRT data were
obtained2 in consolidated yearly files which contain successively the first 24 hours of each
forecast run, starting at 0 UTC. The correspondent model experiment IDs under which
the data can be obtained are f93i for the NRT and fbov , respectively, for the re-analysis.
3.3.2 Aerosol Optical Properties
This section describes the two different sets of aerosol optical properties which were em-
ployed in this study. The OPAC dataset (Optical Properties of Aerosols and Clouds) by
Hess et al. (1998) which is included in the radiative transfer software package libRadtran
was used for feasibility and sensitivity studies. Furthermore an optical property dataset
was generated on the basis of the physical properties of the aerosol types as simulated in
the ECMWF IFS (see Sect. 3.3.1). This dataset allows to derive circumsolar radiation from
IFS output in the most consistent way. Where necessary, OPAC served to complement the
information required for the generation of the optical property dataset, not provided by
ECMWF.
2Data obtained from http://macc.iek.fz-juelich.de/data/f93i/3hourly/ModelLevel/. Data services providedby Forschungszentrum Julich
3.3 Aerosol Input and Processing 35
The OPAC Dataset
OPAC defines so-called aerosol types (e.g. continental-polluted) which represent the ex-
ternal mixture of several aerosol components (e.g. soot or water soluble substances). The
aerosol components discern themselves by their refractive index, density and particle size
distribution. The latter is always parameterized as a logarithmic normal size distribution
(e.g. Gasteiger, 2011, Eq. 2.28)
n(r) =dN
dr=
N0√2π lnσr
exp
[−(
ln r − ln r0√2 lnσ
)2]
(3.5)
where n(r) is the particle number density per radius interval. The parameters of this
distribution are the particle number density N0, the mode radius r0 and the distribution
width σ.
The OPAC optical properties were originally calculated for spheres using the Mie theory
(Mie, 1908). libRadtran contains a modified version in which prolate spheroids represent
mineral dust, instead of spheres. The aspect ratio distribution for the spheroids stems
from microscopic observations made 2006 during the Saharan Mineral Dust Experiment
(Kandler et al., 2011) and the optical properties were modeled as described in Wiegner
et al. (2011), (Pers. comm. Joseph Gasteiger).
Optical Properties for ECMWF IFS Aerosol Components
Optical properties for the aerosol components, which are simulated in the ECMWF IFS,
were created during this study for the use with libRadtran. Thereby it was tried to match
their physical properties as closely as possible. The necessary parameters were taken
from (Morcrette et al., 2009, and pers. comm. Jean-Jacques Morcrette, 2012). However,
information concerning the refractive index could not be confirmed with certainty since
the product is work in progress and not fully documented. The refractive indices used
for the optical properties within the IFS model itself stem from the ADIENT3 project
(pers. comm. Jean-Jacques Morcrette, 2012). However the corresponding project report
(Highwood, 2009) does not always give unambiguous recommendations. In these cases
reasonable assumptions were made.
3http://www.reading.ac.uk/adient/ADIENT.html
36 3. Tools and Methods
For the generation of the optical properties Josef Gasteiger (pers. comm. 2012) provided
an extensive database of Mie calculations. An associated program allows to generate bulk
optical properties according to given size distributions and spectral refractive indices.
Like in OPAC, the aerosol size distributions in the IFS are all based on the log-normal
distribution (Eq. 3.5). The IFS aerosol types are either composed of a one-modal or
a two-modal log-normal distribution. The latter is a superposition of two log-normal
distributions. Furthermore for some aerosol types the distributions are subdivided into
several size bins. The size distribution parameters for the different aerosol types are listed
in Tab. 3.1. With these it is possible to calculate the volume distribution and the total
aerosol volume. With the density ρ, the mass concentration ρV ([g/m3]) can be derived.
The mass concentration is the quantity which is accepted by libRadtran to scale the aerosol
properties accordingly. The density values as well as the source of the used refractive
indices for each aerosol type are listed in Tab. 3.2. In the following details concerning the
individual types are given.
Dust
The size distribution is one-modal with r0 = 0.29 µm, σ = 2 and N0 = 1 cm−3. Since
information on the refractive index could not be determined conclusively, the spectrally
resolved refractive index from OPAC was used. Highwood (2009) discusses different pos-
sible data sources for refractive indices for dust, noting that OPAC shows relatively high
absorption.
Sea Salt
Sea salt is hygroscopic, i.e. it takes up water dependent on the ambient relative humidity
(RH). Correspondingly the aerosol grows. The size distribution is two-modal with [mode]
radii (at 80% RH) of 0.1992 µm and 1.992 µm, σ = 1.9/2.0 and N0 = 70 cm−3/3 cm−3 (pers.
comm. J.-J. Morcrette). The growth factor fg is defined as
fg =r0(RH)
r0(RH = 0%). (3.6)
Growth factors used in OPAC and the IFS were compared. Figure 3.2 shows that they are
similar. In the end the growth factors as well as the refractive index values from OPAC
3.3 Aerosol Input and Processing 37
were used. Highwood (2009) also recommends to use the values from Shettle and Fenn
(1979), which are the basis for the OPAC implementation. With the utilized growth factors
from OPAC the mode radii at RH = 0 % evaluate to 0.1002 µm and 1.002 µm, respectively.
The density of the dry particles was taken from (Morcrette et al., 2009) as 2.16× 106 g/m3
and was reduced according to the mass mixing ratios of salt and water for growing particles
with the density of water being 1.0× 106 g/m3.
Figure 3.2: Hygroscopic growth factors for the sea salt (SS) accumulation mode (AM) andthe coarse mode (CM) of OPAC, as well as for the Sea salt as implemented in the ECMWFIFS (pers. comm. J.-J. Morcrette).
Sulfate, Organic Carbon and Black Carbon
The three aerosol types organic carbon, black carbon and sulfate exhibit the same mode
radius of 0.0355 µm and have the same underlying size distribution with σ = 2 and N0 =
1 cm−3 (pers. comm. J.-J.Morcrette). They only differ in the refractive index.
In the data description tables from the ECMWF homepage4 the organic component is
named “organic matter”. From this name one could assume that this species includes
gas-to-particle components as well brown carbon and pollen et cetera. A variety of optical
properties is therefore thinkable. However, in personal communication J.-J. Morcrette
referred to OC or organic carbon and mentioned that the source for the refractive indices
4http://www.ecmwf.int/publications/manuals/d/gribapi/param/search=organic/ (last accessed03/Jul/2013)
38 3. Tools and Methods
is the ADIENT database. Consequently, I used the refractive indices from ADIENT for
organic carbon. Highwood (2009) describes organic carbon as “humic-like substances”
that “generally derive from fossil fuel burning”. Since this includes many substances,
the refractive indices may vary strongly. ADIENT uses the refractive index for Swannee
River Fulvic Acid (SRFA), which is commonly used as proxy for organic carbon, as anchor
point at 532 nm. From this reference value it employs a wavelength dependency which is
collected from different studies, depending on wavelength interval. The ADIENT report
states that the wavelength dependent refractive indices are given at the OPAC wavelengths.
However, the wavelengths used by J. Gasteiger’s tools are slightly different to the ones in
the files obtainable from the ADIENT homepage. Therefore, the refractive indices were
interpolated linearly to the required wavelengths.
In the IFS the black carbon and the organic carbon component are both divided into a
hydrophilic and hydrophobic part. For the hydrophilic part I used in principle the same
refractive indices as for the hydrophobic part, but mixed with water. I.e. for the hydrophilic
part the mode radius grows with RH and the refractive indices are mixed with the ones for
water from the OPAC database5 according to the volume mixing ratio. Hygroscopic growth
factors from the OPAC component water-soluble (waso) were used for technical reasons for
organic carbon, black carbon and sulfate, which is always hygroscopic. Figure 3.3 compares
them to the growth factors as used originally for the optical properties in the IFS. Alike
the growth factors for sea salt they are similar.
The densities of dry organic and black carbon were both set to 1.769× 106 g/m3 (pers.
comm. J.-J.Morcrette). For sulphate the density of the OPAC component waso was used
(1.80× 106 g/m3) due to lack of other information.
For sulfate as well as for black and organic carbon the size distribution was given by J.-
J. Morcrette (pers. comm. 2012), however he did not state over what range the optical
properties are integrated in the IFS. Hence, the integration was performed from 0.001 µm
to 5 µm to include all possible particle sizes.
Discerning the different size bins and hygroscopic and hydrophilic parts, optical properties
for 11 different aerosol components have been generated in total.
5Originally the refractive indices for water stem from Hale and Querry (1973).
3.3 Aerosol Input and Processing 39
Figure 3.3: Hygroscopic growth factors for the OPAC components “water soluble” (waso)and “sulfate droplets” (suso), as well as for the for IFS aerosol types organic carbon (OC),black carbon (BC) and sulfate (SO4).
Table 3.1: Parameters of the size distributions underlying the aerosol optical propertiesgenerated for the use with IFS aerosol data. For multi-modal size distributions the valuesfor r0, σ and N0 are separated by a slash. All values are valid for RH = 0 %.
Type r0 [µm] σ N0[cm−3] Size bin limits [µm]Dust 0.29 2 1 0.03 / 0.55 / 0.9 / 20Sea Salt 0.1002 / 1.002 1.9 / 2.0 70 / 3 0.03 / 0.50 / 5 / 20Org. Carbon 0.0355 2 1 0.001 / 5Black Carbon 0.0355 2 1 0.001 / 5Sulfate 0.0355 2 1 0.001 / 5
3.3.3 Post-Processing of ECMWF Output
The IFS provides the aerosol mass mixing ratio ζ ([kg kg−1]) for each aerosol component
for the individual model boxes. For the radiative transfer simulations in this study with
libRadtran aerosol was always placed in a layer between 0 km and 1 km with a constant
mass concentration ρV ([g m−3]). Since for circumsolar radiation the integrated aerosol
loading of the atmospheric column is of relevance, the area mass loading ρA ([g/m2])
must be calculated as the common quantity. In case of the libRadtran calculations this is
straightforward as ρA = 1000 m · ρV . For the IFS data the area mass loading is calculated
as ρA =∑N
i=0 ζi · ρA,air,i where N is the number of model layers and ρA,air,i the mass of air
40 3. Tools and Methods
Table 3.2: Density values ρ, as well as the sources of the refractive indices used in thegeneration of optical properties for the IFS aerosol types.
Type ρ(RH = 0 %) Refrac. Index SourceDust 2.61× 106 g/m3 OPACSea Salt 2.16× 106 g/m3 OPACOrg. Carbon 1.77× 106 g/m3 interpol. from ADIENTBlack Carbon 1.77× 106 g/m3 OPAC (soot)Sulfate 1.80× 106 g/m3 OPAC (waso)
per m2 in the layer (i+ 1). ρA,air,i is calculated from the pressure on the model levels i and
i+1 as ρA,air,i = (pi+1−pi)/g0 with the gravitational acceleration g0 (assuming hydrostatic
equilibrium). The pressure on the model levels can be reconstructed from the coefficients
ai and bi, which are tabulated for every model level, and the surface pressure psurf , which is
given for every model column as pi = ai ·1× 105 Pa+ bi ·psurf . ~a, ~b and psurf are distributed
along with the IFS model output.
The IFS output was available 3-hourly. While this seems appropriate to capture the aerosol
concentration changes, the circumsolar radiation is also influenced by the sun zenith angle
because it depends on the slant path optical thickness (Eq. 2.3). Therefore a 3-hourly
resolution is not enough to capture its distinct daily course. Circumsolar radiation values
were therefore calculated every 15 minutes with the aerosol properties being interpolated
linearly from the 3-hourly values.
3.4 Atmospheric Radiative Transfer
3.4.1 The RTE-Solvers DISORT and MYSTIC
When it comes to simulating (unpolarized) radiances considering a horizontally homoge-
neous atmosphere, the application of the discrete ordinate method (e.g. in the implemen-
tation DISORT by Stamnes et al., 1988) is a common choice. It is thoroughly validated
and freely available. However it has two drawbacks that complicate or sometimes even
prohibit the correct simulation of circumsolar radiation. First, DISORT can only deal
with a point source for the incoming radiation at top of atmosphere. For the correct cal-
culation of radiances in the circumsolar region, the sun’s extent has to be considered (see
Sect. 2.5). This could also be achieved with DISORT by performing several simulations,
3.4 Atmospheric Radiative Transfer 41
placing the point source at different spots within the sun disc, and weighting the results
with the extraterrestrial radiance distribution, as suggested by Stamnes et al. (2000). This
procedure, however, would strongly increase the computational time. One could also use
just one simulation and fold the resulting sunshape with the extraterrestrial radiance dis-
tribution. This would take as given that the one radiative transfer simulation performed
for a point source is representative of the whole sun disk – which would probably hold
true for most (one dimensional) simulations required for this study. While the limitation
of DISORT to a point source can therefore be considered merely as inconvenience, it is a
real problem that it cannot reproduce the extreme forward peaks of some scattering phase
functions: In the discrete ordinate method the radiative transfer equation is split into an
even number nstr of independent integro-differential equations, called streams. The scat-
tering phase function is expanded in a series of Legendre polynomials of equal length. Only
a limited amount of Legendre polynomials/streams can be used, else the series expansion
gets numerically unstable and the DISORT algorithm computationally inefficient (scales
approximately with n3str, Stamnes et al., 2000). With the limited amount of polynomials,
the forward peaked phase functions of cirrus clouds cannot always be represented well.
Buras et al. (2011) showed, as example, the Legendre series representation of the phase
function of the “Baum v2” optical properties at reff = 60 µm and λ = 500 nm. Even with
20 000 Legendre polynomials – which for this case is about the maximum number before
the series expansion gets instable – the deviations from the original phase function still
reach a few percent. To alleviate these downsides, one normally uses a truncated Leg-
endre series – corresponding to a less forward peaked phase function – but applies a so
called “intensity correction” on the DISORT results (Buras et al., 2011). The intensity
correction however, also shows instabilities for extremely forward peaked phase functions
for certain setups (pers. com. Robert Buras, 2013). Furthermore DISORT does not cope
well with some geometries. For example, with the sun in the zenith, the result must be
extrapolated from the next zenithal supporting point, which depends on nstr. In situations
with forward peaked phase functions this can lead to erroneous results, as shown further
down in this section. Because of these deficiencies, DISORT was not used in this study to
simulate radiances in the vicinity of the sun. However, the lookup tables for APICS, for
which the forward peak of the scattering phase functions as well as the extent of the sun
play no important role, have been computed with the C language version of DISORT from
libRadtran, called CDISORT (Buras et al., 2011), using 16 streams.
MYSTIC (Emde and Mayer, 2007; Mayer, 2009; Emde et al., 2011; Buras and Mayer,
42 3. Tools and Methods
2011) – which stands for “Monte Carlo code for the physically correct tracing of photons
in cloudy atmospheres” – is the Monte Carlo solver from the libRadtran radiative transfer
package (Mayer and Kylling, 2005). Its principle to solve the radiative transfer is to trace
individual photons through the atmosphere. All events in which the photon interacts
with the atmosphere like scattering and absorption are treated statistically according to
their physical probability density functions. Forward peaked scattering phase functions
may also pose a problem for plain Monte Carlo simulations, as they may cause rare but
result-dominating events – so called “spikes“. Leveling the spikes will increase computing
time excessively. A solution for this problem was given by Buras and Mayer (2011). The
variance reduction methods described by them are implemented in MYSTIC. It therefore
brings along an important prerequisite to simulate the radiance distribution in the aureole
region.
Figure 3.4 compares sunshapes simulated with DISORT and MYSTIC for a point source.
It shows that both RTE solvers produce the same sunshapes for a setup with OPAC dust
aerosols. Although the dust aerosol contains relatively large aerosol particles, the results
agree, even for a low number of streams (nstr = 16). The relative difference compared to
MYSTIC stays below 1% for any of the shown CDISORT simulations, and is therefore in the
range of the Monte Carlo noise of the MYSTIC simulations. In the example with a cirrus
cloud, it seems that CDISORT reproduces the MYSTIC sunshape well for θ approximately
larger than 0.25 but differences occur closer to the source. However, from Fig 3.5, which
shows the relative deviation of the CDISORT results from the MYSTIC results, it becomes
obvious that errors in the magnitude of 20% and even more must be expected in the
circumsolar region. This emphasizes the just discussed drawbacks of DISORT concerning
strongly forward peaked phase functions.
3.4.2 Adaptation of the RTE-Solver MYSTIC
“Backward” Simulation of Aureole Irradiance
The most obvious way to simulate radiance with a Monte Carlo model is to inject photons
at top of the atmosphere and trace them until they are absorbed, back-scattered into space
or eventually reach a detector surface through a finite cone where they are counted. The
probability for photons to enter the detector cone however is vanishingly small. Therefore
most simulated photons will be absorbed or leave the system before reaching the detector
3.4 Atmospheric Radiative Transfer 43
0.0 0.5 1.0 1.5 2.0θ / deg.
103
104
105
106
107
Radia
nce
/ m
W/m
2/n
m/s
r
MYSTIC
nstr = 16
nstr = 64
nstr = 126
0.0 0.5 1.0 1.5 2.0θ / deg.
103
104
105
Radia
nce
/ m
W/m
2/n
m/s
r
nstr = 16
nstr = 32
nstr = 64
MYSTIC
Figure 3.4: Simulations of sunshapes with libRadtran’s solver CDISORT using differentnumbers of streams compared to a simulation with the MYSTIC solver for desert dustaerosol from OPAC (left panel) and a cirrus cloud composed of HEY solid columns (rightpanel). θsun = 0. See Fig. 3.5 for relative deviations between CDISORT and MYSTIC.
0.0 0.5 1.0 1.5 2.0θ / deg.
80
60
40
20
0
20
40
Rela
tive d
evia
tion /
%
nstr = 16
nstr = 64
nstr = 126
0.0 0.5 1.0 1.5 2.0θ / deg.
80
60
40
20
0
20
40
Rela
tive d
evia
tion /
%
nstr = 16
nstr = 32
nstr = 64
Figure 3.5: Relative deviation of the sunshapes simulated with CDISORT to the simulationwith the MYSTIC solver as in Fig. 3.4.
44 3. Tools and Methods
and thus will not contribute to the result. Simulations therefore converge extremely slow
with this method. If the considered radiative transfer problem is one dimensional, that
is, if no horizontal variation of the atmosphere is allowed, the results are horizontally
invariant. The lateral position of the detector is therefore irrelevant and only its directional
orientation matters. Or in other words, one can shorten the simulation time by introducing
new detectors at suitable positions. By performing so called “local estimates” (LEs) the
convergence of the results can be sped up even more. In this case a LE is executed at
every scattering event. That is, the probability is calculated that the photon would be
scattered directly into a detector of the considered directional orientation and reach this
detector without any further scattering or being absorbed. Therefore every scattering
photon contributes to the result. Figure 3.6 illustrates this approach which enhances the
convergence of the simulation considerably.
TOA
LE LE LE
Cloud
Detector
Figure 3.6: Scheme of a forward Monte Carlo calculation of radiance for a one-dimensionalradiative transfer problem utilizing local estimates (LEs, dashed lines). Photons are in-jected at top of atmosphere (TOA). A local estimate is executed at every scattering eventalong the photon path (solid line).
If a three dimensional radiative transfer problem has to be solved, the orientation and
position of the detector must be considered. Tracing photons forward is therefore not
efficient because performing local estimates is not possible anymore. In this case radiances
are usually calculated backward. Thereby the reciprocity principle is used. Photons are
3.4 Atmospheric Radiative Transfer 45
started from the detector in the direction for which the radiance shall be calculated. The
photons are then traced backward through the atmosphere. Again a local estimate is
executed at every scattering event. However, instead of the probability for the photons
being scattered into the direction of the detector, the probability for a scattering into the
direction of the sun is calculated (see also illustration in Fig. 3.8 further down). This is
called the “backward” Monte Carlo method. For this study only 1D-radiative transfer
capabilities are necessary, for which the forward method could have been used. In the end
however, MYSTIC’s backward code was adapted for the accurate simulation of circumsolar
radiation. Such, circumsolar radiation can also be simulated for 3D-scenes, in case future
applications require so.
Considering the calculation of the CSR the integrated value of the radiance over a solid
angle section around the sun is desired. Therefore, I modified the backward method such
that the photons are not only emitted into one single direction but evenly distributed over
the solid angle covered by the detector. The goal is to compute a mean radiance that only
needs to be multiplied by the solid angle to obtain the corresponding irradiance. Note
that by doing so, the cosine term in the integral∫∫
L(α, φ) cos(α) sin(α) dαdφ is safely
neglected (comp. Sect. 2.6).
Despite the variance reduction methods by Buras and Mayer (2011), the backward simula-
tion of mean radiance for a detector with an extended field of view which may suffer from
an unbalanced contribution of the individual photons. This happens if the field of view is
centered at the sun and a strongly forward-peaked scattering phase function is present. In
this case the photons emitted close to the edge of the detector’s field of view will contribute
considerably less to the result than the ones emitted near the center. This is because dur-
ing the first local estimate the outer photons will have a much smaller probability of being
scattered into the direction of the sun than the inner photons: The inner photons need
to be scattered by a smaller angle than the outer ones. Due to the strong gradient in the
forward peak of the scattering phase function even small differences in scattering angle will
lead to large differences in scattering probability. Therefore, on average, the local estimates
for the inner photons will be quite high compared to the ones for the outer photons, and
so will be the contribution to the result. This will slow down the convergence of the result.
To overcome this, I modified the initial distribution of photon directions in MYSTIC, so
that more photons are launched near the center of the field of view than geometrically
justified. This change of sampling strategy must be corrected for by weighting the photon
contributions to the result accordingly. In the end more photons are launched closely to
46 3. Tools and Methods
the sun with lower photon weights and less photons are launched in the outer areas of the
aureole but with increased photon weights which leads to a more balanced contribution of
the individual photons to the result and thus to a faster convergence.
In principle the Monte Carlo method allows an arbitrary sampling strategy as long as it is
accounted for by appropriate photon weights. To initially distribute the photon directions,
I adhered to the function Pddis, which is dynamically generated within MYSTIC from the
scattering phase functions utilized in the simulation. It has already proved useful in the
other variance reduction methods implemented in MYSTIC. Pddis is defined as (Buras and
Mayer, 2011, Eq. 10),
Pddis(θ) = cddis maxi
[Pi(θ)] for all θ ∈ [0, 180] . (3.7)
It is the maximum of all scattering phase functions Pi in the current simulation (e.g. for
different cloud or aerosol types present in the scene). That is, for every scattering angle
θ the maximum value of all scattering phase functions is taken. The resulting function
is then normalized with the coefficient cddis. The initial photon directions are distributed
according to this function which is evaluated in the angular interval covered by the field of
view of the considered detector. Technically this is realized by first directing every photon
towards the sun center and then “scattering” the photon – still at the original position –
according to the virtual scattering phase function Pddis. Thereby only scattering angles are
allowed which are in line with the desired detector geometry. This is achieved by cutting of
Pddis at the angular radius of the detector’s field of view and re-normalizing it afterwards, so
that the total scattering probability remains unity. Figure 3.7 illustrates how this approach
balances the contributions of the individual photons: In the left panel the initial directions
are evenly distributed over the field of view, but most photons contribute only negligibly to
the result. In the right panel the photons are distributed according to Pddis and weighted
accordingly. In this case the photon contribution is much better balanced. Note that both
simulations converge to the same result, but the simulation relying on Pddis for the photon
distribution reaches a better uncertainty level. Expressed as standard deviation it is a
factor seven lower than for the simulation relying on the original photon distribution.
Extraterrestrial Radiance Distribution
To simplify the radiative transfer the sun is commonly assumed to be a point source at
infinity. All sun rays are therefore parallel and enter the atmosphere under the same angle.
3.4 Atmospheric Radiative Transfer 47
0.00000.00080.00160.00240.00320.00400.00480.00560.00640.0072
0.00.10.20.30.40.50.60.70.80.91.0
Figure 3.7: Initial distribution of photon directions for MYSTIC simulations of a detectorwith 2.5 aperture and a scene with a cirrus composed of HEY aggregates (reff = 30 µm,τ = 1.0). The contribution to the result (in arbitrary units) from the first local estimate iscolor-coded. In the left panel the photons are distributed evenly with solid angle. In theright panel the photons are distributed according to Pddis (Eq. 3.7) and their contributionis weighted accordingly to this modified distribution. Note the different color scales.
However this approximation is not appropriate when simulating radiance in the vicinity
of the sun disk. The sun has a finite angular extent and the center of the sun disk is
brighter than the limb. This so called limb darkening is caused by absorption in the sun’s
atmosphere and is therefore wavelength dependent. To take account of these effects I
implemented a disk source in MYSTIC which uses a wavelength dependent limb darkening
model. The radiance distribution Lλ relative to the radiance from the center of the sun
L0,λ is given in (Scheffler and Elsasser, 1990) and (Kopke et al., 2001) as
LλL0,λ
=1 + βλ cos θ
1 + βλ=
1 + βλ√
1− r2
1 + βλ. (3.8)
Here θ denotes the exit angle of the radiance normal to the surface of the sun. For our
purpose the second formulation is more practical where r = 0 represents the center of the
sun disk and r = 1 the limb; simple geometric considerations on the unit circle confirm
that√
1− r2 = cos θ. The limb darkening coefficient βλ depends on the wavelength; I
adopted the corresponding formula from Kopke et al. (2001) (originally from Waldmeier,
1941):
βλ =3hc 4√
2
8kλTs(3.9)
48 3. Tools and Methods
TOA
a
a
a
a
a
b
b
b
b
Detector
Clouds
Figure 3.8: Illustration of the “backward” Monte Carlo method for the simulation ofradiance. Photons are started from the detector and traced until they are absorbed orleave the atmosphere. Two photon paths (solid lines) are depicted (a & b). The initialphoton directions are distributed within the detector cone. At each scattering event alocal estimate (dashed lines) is performed which contributes to the result. A local estimatedirection on the sun disk is randomly assigned to each photon with a probability accordingto the limb darkening (Eq. 3.10). This LE-direction is fixed throughout the photon journey.
3.4 Atmospheric Radiative Transfer 49
where λ is the wavelength in m, h the Planck constant (6.6261× 10−34 J s), c the speed
of light (2.9979× 108 m s−1), k the Boltzmann constant (1.3806× 10−23 J K−1) and Ts the
temperature of the sun’s surface (5740 K).
A disk source is used in the simulation of the diffuse as well as of the direct radiation.
Since photons are traced backward, the disk source has to be considered in the local
estimate direction. Each photon gets randomly assigned one direction, into which all LEs
are performed. The LE directions are sampled on the sun disk according to the limb
darkening function. That is the probability density p for a photon having assigned a LE
direction pointing to a spot on the sun disk with angular distance α from center is given
as
p =
Lλ( α
αsun) sinα
αsun∫0
Lλ( ααsun
) sinα dαfor α ≤ αsun
0 else .
(3.10)
The sine term accounts for the changing solid angles covered by annular sections of constant
dα as α changes. Figure 3.8 illustrates the backward tracing of photons with a sun disk as
source.
In case direct radiation is to be included in the simulation, it is checked directly after
the generation of each photon whether the randomly assigned local estimate direction is
covered by the sensor’s field of view. If so, the probability of the photon reaching the sun
without scattering or absorption is calculated in analogy to a LE to obtain the contribution
of the direct radiation to the result. This is performed even before the photon starts to
move. Only then the photon is launched on its journey through the atmosphere.
The angular sun radius αsun can be set manually or be determined automatically if a time
and date is given. The simulations performed for this study, however, neglect the course
of the year of αsun and refer to a mean angular sun radius of 0.266. From reference
simulations with varying αsun it is estimated that the absolute error in CSR that arises
from this is less than ±0.004 as long as αcir > 0.375. However, as Neumann et al. (2002)
have demonstrated, the actual sun radius should be taken into account when calculating
CSR from ground based sunshape measurements.
The just discussed model improvements can be followed in Fig. 3.9. The differences between
a simulation of the diffuse radiance with a point source (blue) and a disk source (green) are
pronounced for angles smaller than ≈ 1 from the sun center which is in line with findings
by Grassl (1971). The red curve shows the total sunshape consisting of direct and diffuse
50 3. Tools and Methods
radiation.
Figures 3.10 and 3.11 demonstrate the capabilities of MYSTIC to simulate the radiative
transfer for a 3D scene. They are intended to give the reader an idea of how the circumsolar
radiation caused by cirrus clouds and aerosols appear to the human eye. The simulations
were performed with the option “output RGB”. This prompts MYSTIC to calculate ra-
diances at several wavelengths between 380 nm and 780 nm and to weight these results
into values for red, green and blue color similar to the perception of the human eye. For
the simulations concerning aerosol (Fig. 3.11) some cumulus clouds were introduced to the
scene to provide more contrast. The corresponding 3D cloud field was produced by Katrin
Wapler performing a large eddy simulation (LES). For detail on the LES refer to Wapler
(2007).
Figure 3.9: Simulated broad band sunshape for a cirrus cloud with optical thickness 0.5and with the sun in the zenith. Blue: Diffuse radiance for a point source. Green: Diffuseradiance for a disk source. Red: Direct and diffuse radiance for a disk source.
3.4.3 Pseudo-Spectral and Solar Integrated Radiative Transfer
As mentioned, the RTE (Eq. 2.4) must be solved for each wavelength individually. If RT
simulation results are given at a certain wavelength in this study, they were obtained by
a pseudo-spectral calculation treating the molecular absorption with the band parameter-
ization from LOWTRAN as implemented in libRadtran (adapted from Ricchiazzi et al.,
1998). Where the results are “broad band” (bb) or solar/shortwave integrated, they have
been obtained using the kato2 correlated-k distribution as provided by libRadtran. kato2
3.4 Atmospheric Radiative Transfer 51
Figure 3.10: MYSTIC RGB simulations of the diffuse radiation, Albedo = 0.2, θsun = 49,Viewing elevation: −17.5 till +44.4. Azimuthal field of view: 80, Left: Pure Rayleighatmosphere. Right: Inhomogeneous cirrus layer with a maximum optical thickness of≈ 0.35.
Figure 3.11: MYSTIC RGB simulations of the diffuse radiation, Albedo = 0.2, θsun =49, Viewing elevation: −17.5 till +44.4. Azimuthal field of view: 80, Left: Rayleighatmosphere with fair weather cumuli. Right: Additionally with a layer of aerosol (OPACdesert dust, aerosol optical depth 0.5). Cumulus clouds by K. Wapler.
52 3. Tools and Methods
is an update of the version described by Kato et al. (1999). It uses optimized tables pro-
vided by Seiji Kato which allow to reach similar precision with a reduced number of 148
sub-bands instead of 575, which corresponds to a reduced number of RTE-solver calls.6
3.5 Parametrization of Circumsolar Radiation
The Monte Carlo radiative transfer solver MYSTIC was adapted during this study such
that it can simulate the radiance distribution in the circumsolar region with high accuracy
(Sect. 3.4.2). However, these simulations are computationally too expensive to perform
them by the number of millions needed to evaluate the satellite or weather model data
used in this study. Therefore in this section a way of efficiently parameterizing the circum-
solar radiation – and especially the CSR – is developed. A table of coefficients obtained
from MYSTIC simulations together with the concept of an apparent optical thickness by
Shiobara and Asano (1994) acts as a basis for this. Parts of this section are adapted form
Reinhardt et al. (2013).
As mentioned, the circumsolar radiation relevant for CST applications is mainly caused by
forward scattering within aerosol or thin cirrus layers. When parameterizing circumsolar
radiation it is sufficient to focus on the properties of the atmospheric layers containing
scattering particulates. Therefore, varying Rayleigh scattering due to changing sun zenith
angle θsun or different elevations as well as surface albedo changes are neglected in the
following. The effects of these simplifications were analysed by several simulations for the
largest and therefore most sensitive field of view considered in this study, αcir = 5. The
control simulations show that Rayleigh scattering in a pure atmosphere causes a shortwave
integrated circumsolar irradiance of less than 1 W/m2 and a CSR of less than 0.0015 even
for the extreme assumption of the surface albedo being 1. This was tested for varying sun
zenith angles θsun between 0 and 88. To assess the effect of a changing surface albedo, it
was varied between 0 and 1 considering different ice clouds (0 < τ < 3). Resulting changes
in diffuse shortwave integrated irradiance were always below 2 W/m2 or 0.0025 in CSR
respectively. Therefore, unless stated otherwise, the simulations used for the development
of the circumsolar radiation parameterization were performed with θsun = 0, albedo 0 and
elevation 0 m.7
6See also the libRadtran user manual.7For aerosol a θsun dependence for high sun zenith angles was diagnosed and accounted for (Sect. 3.5.3).
3.5 Parametrization of Circumsolar Radiation 53
The calculation of circumsolar radiation in this study is limited to sun zenith angles smaller
than 80. For larger sun zenith angles the plane parallel assumption introduces consid-
erable errors. Furthermore also the cloud property retrieval is not reliable for θsun > 80
(Sect. 3.2). However this limitation is of little importance as many CST facilities are inop-
erative at such low solar altitude due to the rapid decrease of direct irradiance for larger
sun zenith angles and due to the increased self shadowing.
3.5.1 Concept of the Apparent Optical Thickness
Shiobara and Asano (1994) and Kinne et al. (1997) corrected sun photometer measurements
in case of cirrus clouds with the concept of an apparent optical thickness. Recently Segal-
Rosenheimer et al. (2013) developed this approach further into a cloud property retrieval for
sun photometer data. In the following the concept of an apparent optical thickness serves
to parameterize circumsolar radiation. We shall see that this approach enables to calculate
circumsolar radiation elegantly and in a flexible way from a comparatively low number of
pre-calculated coefficients. The concept allows to eliminate the optical thickness of the
scattering layer as degree of freedom. Furthermore, circumsolar radiation in the presence
of several different scattering layers can easily be calculated because the apparent optical
thickness can be assumed to be additive without making a large error. In the following the
concept is outlined.
The direct transmission T through the atmosphere can be decomposed into a particulate
and molecular transmission
T = TpTm (3.11)
The particulate transmission Tp is expressed as
Tp = exp(−τs) (3.12)
where τs is the particulate slant path optical thickness along the line of sight from the
observer to the sun. The molecular transmission Tm is determined by Rayleigh scattering
and absorption by air molecules.
Diffuse radiation – that is radiation having been scattered in the atmosphere – will enter
any optics with a finite field of view pointed towards the sun in addition to the direct
54 3. Tools and Methods
radiation. The direct radiation has not been scattered and therefore stems only from the
sun while diffuse radiation can come from both – the sun disk region and the circumsolar
region.
Considering the total radiation entering the field of view one may consider an apparent
transmission T ′ which describes both, the diffuse and the direct contribution. T ′ can as
well be decomposed into a particulate and molecular part.
T ′ = T ′pT′m (3.13)
Since Rayleigh scattering on molecules contributes only a negligible part to the radiation
in the circumsolar region, one can approximate T ′m = Tm. The molecular transmission Tm
will not be further discussed here since it will cancel out later in the relevant formulas.
The apparent particulate transmission T ′p can be parameterized as
T ′p = exp(−kτs) , (3.14)
with k taking values between 0 and 1. This means that the difference between the direct
particulate transmission – following Beer’s law – and the apparent particulate transmission
can be accounted for with the factor k. Defining the apparent optical thickness τapp one
obtains
τapp = kτs , (3.15)
T ′p = exp(−τapp). (3.16)
It is notable that the corrective factor k depends mainly on the field of view, reff and the
ice particle shape or the aerosol type but is almost independent of τs itself. This holds true
as long as the optical thickness does not get too large. Extensive Monte Carlo simulations
performed during this study with different ice particle types as well as with different aerosol
species showed that as long as 0 < τs < 3, k varies by less than 3% if all other parameters
except the optical thickness are kept constant.
A lookup table approach allows the fast computation of CSR using a parameterization in-
stead of solving the radiative transfer equation. Thereby k is interpolated linearly between
pre-tabulated values obtained from MYSTIC simulations. The corresponding ansatz is
outlined in the following.
3.5 Parametrization of Circumsolar Radiation 55
If we denote the circumsolar irradiance entering an aperture with an opening angle α as
Icir, the total irradiance entering the same aperture as Itot,α and the total irradiance coming
solely from the sun disk as Itot,sun, then we can write the circumsolar ratio as
CSR =Icir
Itot,sun + Icir
=Itot,sun + Icir − Itot,sun
Itot,sun + Icir
= 1− Itot,sun
Icir + Itot,sun
= 1− Itot,sun
Itot,α
. (3.17)
In general we can express the total irradiance Itot,sun and Itot,α for a given atmosphere by
Itot,sun = I0T′p, sun = I0 exp(−ksunτs) (3.18)
Itot,α = I0T′p, α = I0 exp(−kατs) (3.19)
with ksun being the corrective factor for a field of view correspondent to the sun’s angular
radius and kα for a field of view correspondent to the limiting angle for which the CSR
shall be calculated. I0 denotes the solar constant Is corrected for molecular transmission:
I0 = IsTm . (3.20)
Applying Eqs. (3.18) & (3.19) to Eq. (3.17) yields
CSR = 1− exp(−ksunτs)
exp(−kατs)= 1− exp[−(ksun − kα)τs] . (3.21)
One should note that ksun and kα are subtracted which can lead to an adding of the
individual errors. Therefore, the error in CSR due to tabulating them independently of
the optical thickness can reach up to 20%, but most times it is well below 10%. If higher
precision is required, k can be tabulated at several supporting points in τ .
Since the angular radius of the sun is assumed constant, we can re-write Eq. (3.21) as
CSR = 1− exp(−∆kατs) (3.22)
with ∆kα = ksun − kα. One can even simplify further:
CSR ≈ ∆kατs (3.23)
56 3. Tools and Methods
as long as ∆kατs is much smaller than 1. This is often the case for aerosol for which, in
contrast to cirrus clouds, both ∆kα as well as τs are often small. Figure 3.12 compares the
exponential ansatz and the linear approximation exemplary for one type of cirrus cloud
and one type of aerosol. For the considered aerosol type the linear approximation can be
used even for high values of τs without making a large error. For cirrus clouds, however,
considerable errors arise for τs ' 0.5.
0.0 0.5 1.0 1.5 2.0τ550nm
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
CSR
(2
.5)
Cirrus
Cirrus lin. approx.
Aerosol
Aerosol lin. approx.
Figure 3.12: CSR for a cirrus cloud (Baum v2.0, reff = 25 µm) and for aerosol (OPAC min-eral dust accumulation mode) calculated once with the precise exponential ansatz (Eq. 3.22,solid lines) and once with the linear approximation (Eq. 3.23, dashed lines).
For cirrus clouds I tabulated broad band values of the corrective factor k as a function of
three parameters: The field of view which is characterized by the aperture’s opening half
angle8 α, the particle shape and the particle size9 reff , Please note that APICS returns the
cirrus optical thickness at 550 nm, however for this study the integrated solar broad band
(bb) circumsolar radiation is in the focus. Therefore, the conversion from optical thickness
at 550 nm to the integrated solar value is also incorporated into k. Hence the tabulated
values of k discussed in the following will translate a slant path optical thickness at 550 nm
into a broad band apparent optical thickness. The broad band optical thickness of cirrus
clouds does in general not differ much from the optical thickness at 550 nm. However,
because this small conversion factor is included, k can at times be slightly larger than 1.
8supporting points 0.266, 0.375, 0.5, 0.75, 1.0, 1.5, 2.0, 2.5, 3.0, 4.0 and 5.09supporting points 5 µm, 10 µm, 15 µm, 20 µm, 25 µm, 30 µm, 40 µm, 50 µm, 60 µm, 70 µm, 80 µm and 90 µm,as far as covered by the optical properties.
3.5 Parametrization of Circumsolar Radiation 57
Table 3.3: Exemplary values of k (Eq. 3.14) for varying ice optical properties.
Optical Properties k(0.266) k(2.5) k(5.0)Baum v2.0, reff = 10 µm 0.97 0.65 0.55Baum v2.0, reff = 25 µm 0.82 0.46 0.43Baum v2.0, reff = 60 µm 0.52 0.33 0.32Baum v3.5, reff = 10 µm 0.96 0.60 0.53Baum v3.5, reff = 25 µm 0.80 0.52 0.49Baum v3.5, reff = 60 µm 0.65 0.50 0.47
To obtain k, the irradiance Itot within the considered field of view was simulated with
MYSTIC for clouds with an optical thickness at 550 nm of 1.5. k is then calculated as
k =log(
I0Itot
)1.5
. (3.24)
I0 is obtained by a MYSTIC simulation of the irradiance within a 0.266 field of view for
a cloud- and aerosol-free atmosphere. An exemplary excerpt of the lookup table for cirrus
clouds is given in Tab. 3.3.
For aerosols the parameters aerosol type/component, field of view and relative humidity10
were considered as variables in the tabulation of k. The ECMWF IFS primarily delivers
(dry) aerosol mass mixing ratios which are converted into area mass loadings. Therefore
k was tabulated in a different way for aerosol than for clouds; for aerosol k converts from
area mass loading directly to apparent optical thickness:
τapp = kaerosolρAµ
. (3.25)
It should be noted that the outlined concept can serve in the derivation of other circumsolar
radiation parameters besides CSR as well. The diffuse irradiance in the circumsolar region
Icir = Itot,α− Itot,sun for example is the relevant parameter when considering solar resource
overestimation by pyrheliometers. Since Itot,α is basically the integral of the sunshape L
(comp. denominator in Eq. (2.13)) over the solid angle, one can also obtain the mean value
10supporting points 0 %, 50 %, 70 %, 80 %, 90 %, 95 %, 98 % and 99 %
58 3. Tools and Methods
of the sunshape L between the limiting angles α1 and α2 by numerical differentiation of
Itot,α with respect to the solid angle Ω enclosed by the corresponding field of view as
L(α1 < α < α2) =Itot,α2 − Itot,α1
Ω(α2)− Ω(α1). (3.26)
This way the sunshape can be coarsely reproduced from the k-LUT, although this may be
not the most straightforward way of use. An example for this is shown in Fig. 3.13. The
leaps in the reconstructed sunshape occur at angular supporting points of the k lookup
table, in between them k is interpolated linearly.
Figure 3.13: Broad band (300 nm – 2600 nm) sunshapes for a cirrus with τs = 0.5,reff = 40 µm (HEY solid-columns). Blue: From a MYSTIC simulation. Green: Calcu-lated according to Eq. (3.26) from the lookup table for k.
For all applications of the k-LUT for which I0 does not cancel out (e.g. calculation of
Icir) and is required as function of the sun zenith angle, I0 was interpolated linearly from
simulations of 1 resolution in θsun.
Figure 3.14 summarizes the basic concept of the outlined parameterization in a flowchart.
3.5.2 Treatment of Multiple Scattering Layers – The Adding
Method
So far the developed parameterization allows to calculate circumsolar radiation caused by
a single layer of scattering particles only. However at times more than one layer has to be
3.5 Parametrization of Circumsolar Radiation 59
SEVIRI MeasurementR(0.6 µm), R(1.6 µm)
APICScloud property retrieval
-> effective radius-> optical thickness
ECMWF IFSaerosol concentration
Circumsolar radiation due tocirrus clouds or aerosol
LUT-assissted parameterizationbased on concept of apparent
optical thickness
A priori InformationGround Albedo
Cirrus Cloud Mask
Figure 3.14: Flowchart of the developed method to derive circumsolar radiation.
60 3. Tools and Methods
dealt with; be it in the case of a cirrus above an aerosol layer or when an external mixture11
of aerosols is to be considered, which can be regarded as multiple individual layers as well.
The latter is the case if IFS data is to be evaluated, because the model output is an
external mixture of aerosol components. The optical thickness of several scattering layers
is additive, and since the parameterization is based on the apparent optical thickness, it
was obvious to test whether it is additive as well.
For this two different test scenarios were considered. The first one places cirrus clouds over
aerosol layers, the second one deals with external aerosol mixtures. In both cases reference
simulations were performed with MYSTIC, treating the multiple layers explicitly. The
sum of the parameterized values of apparent optical thickness for the individual layers
computed according to Eq. 3.15 were then compared to these simulations. The summation
of τapp values of individual layers or components is called “adding method” in the following.
The tests shown in the following were performed for a field of view of 3.0. The standard
deviation in τapp,bb derived from the MYSTIC reference simulations due to Monte Carlo
noise is always smaller than 0.8%.
The cirrus clouds in the first test scenario were always composed of HEY solid-columns.
The ice optical thickness at 550 nm was varied between the three values of 0.1, 0.5 and
1.2 and the effective radius was set to 5 µm, 10 µm, 20 µm, 30 µm, 40 µm, 50 µm, 70 µm or
90 µm. For the aerosol layer optical properties for the three different size bins of the IFS
dust aerosol were variantly used at aerosol optical thickness values at 550 nm of 0.1, 0.3
and 1.2. In total 3 · 8 · 3 · 3 = 216 scenes were created. Figure 3.15 compares the resulting
broad band apparent optical thickness values as well as the resulting CSR values. The
error in the apparent optical thickness due to simply adding values of two layers instead of
explicitly simulating it stays below 4%. All cases with errors > 1.1 % employ aerosol with
the highest AOT considered of 1.2. The error in the CSR is in general higher than in the
apparent optical thickness since for its calculation two τapp-values need to be determined
(comp. Eq. 3.21) which are both prone to errors. The adding of apparent optical thickness
values seems by trend to introduce a negative bias in the CSR since for over 90% of the
considered test cases the parameterized CSR is smaller than the MYSTIC reference value.
For the second test scenario several external aerosol mixtures were considered. Again
the broad band apparent optical thickness and the CSR were calculated – once explicitly
with MYSTIC and once by adding the parametrized apparent optical thickness values for
11External mixture means that particles of different type are present, whereas internal mixture indicatesthat individual particles are composed of several components (e.g. particles with a coating).
3.5 Parametrization of Circumsolar Radiation 61
the individual aerosol species. Overall 230 scenes with a total aerosol optical thickness
at 550 nm of either 0.05, 0.10, 0.15, 0.3, 0.4, 0.5, 0.7, 0.9, 1.1, 1.5, 2.0, 3.0 or 5.0 were
generated by randomly mixing eight aerosol components from OPAC, namely insoluble
(inso), water-soluble (waso), soot, mineral nucleation mode (minm), mineral accumulation
mode (miam), mineral coarse mode (micm), sea salt accumulation mode (ssam) and sea
salt coarse mode (sscm). That is, after randomly picking one total aerosol optical thickness
level τ , each aerosol component i was assigned a random number ti ∈ [0, 1], from which
the partial aerosol optical thickness values τi were computed as
τi = τti∑
aer. components
ti. (3.27)
Figure 3.16 shows that the relative error in the apparent optical thickness is mostly below
2%. Only for aerosol optical thickness values larger than 1 the error can reach values
in the order of 10%. Like in the first test case the errors in CSR can be larger than in
apparent optical thickness. At times they reach values in the order of 15%. Below a total
aerosol optical thickness of ≈ 0.5 the “adding method” underestimates the CSR values from
MYSTIC in tendency while for larger values mostly an overestimation can be observed.
Considering the results of the two test cases I assume that the error in CSR due to ap-
plication of the “adding method” is on average below 5%. For individual setups the tests
showed errors in the range of up to 15%. These errors seem to be acceptable if one consid-
ers the greatly enhanced flexibility obtained by the adding method. Note that if linearity
can be assumed, i.e. ∆kaτs 1 (comp. Eq. 3.23), also the CSR is additive and CSR
contributions of the individual layers can be separated.
62 3. Tools and Methods
Figure 3.15: Values of apparent optical thickness (left) and CSR (right) for 216 setups of atwo layer scene with a cirrus over an aerosol layer. The cirrus is composed of HEY “solid-columns” of varying reff and an ice optical thickness at 550 nm of 0.1, 0.5 or 1.2. The aerosollayer is composed either of the small, medium or large dust component of the ECMWFIFS aerosol (Sect. 3.3.2) with an aerosol optical thickness of 0.1, 0.3 or 1.2. The diagramsare primarily sorted by increasing combined optical thickness (aerosol + cirrus) at 550 nm(τ550nm) and secondly by apparent optical thickness. Upper left: Broad band apparentoptical thickness explicitly simulated with MYSTIC (blue), apparent optical thickness (bb)obtained by adding apparent optical thickness values of the individual layers computedusing the k-LUTs (Sect. 3.5, green), combined optical thickness τ550nm (red). Lower left:Relative deviation of the parametrized apparent optical thickness to the MYSTIC reference.Upper right: Resulting CSR values. Lower right: Relative deviation of the parametrizedCSR to the MYSTIC reference.
3.5 Parametrization of Circumsolar Radiation 63
Figure 3.16: Values of apparent optical thickness (left) and CSR (right) for 230 setups witha random external mixture of eight OPAC aerosol components (inso, waso, soot, minm,miam, micm, ssam, sscm). The diagrams are primarily sorted by increasing combinedoptical thickness which was set to either 0.05, 0.10, 0.15, 0.3, 0.4, 0.5, 0.7, 0.9, 1.1, 1.5,2.0, 3.0 or 5.0 and secondly by apparent optical thickness. See Fig. 3.15 for description ofpanels.
64 3. Tools and Methods
3.5.3 Sensitivity of the CSR Parameterization on Sun Zenith
Angle
The developed CSR parameterization accounts only indirectly for the sun zenith angle
θsun. It is considered in the conversion from optical thickness to slant path optical thick-
ness τs = τcos(θsun)
. However, the MYSTIC simulations to determine k were always per-
formed with θsun = 0. In the following the impact of this simplification is evaluated
with additional simulations. To this end, CSR values calculated from simulations at
θsun = 0 were compared to CSR values at θsun 6= 0 but for equal slant path optical
thickness values. Limiting angles αcir of 0.5, 1.0, 2.0, 3.0 and 5.0 were considered.
The simulations for cirrus clouds were performed with HEY solid-column particles at
reff = 5 µm, 20 µm, 40 µm and 90 µm and τs varying between 0.2 and 4.0. Simulations
for aerosol were performed for the aerosol components miam and ssam from OPAC as well
as the different dust and sea salt components from the IFS with τs varying between 0.1
and 3.0.
0 10 20 30 40 50 60 70 80θsun /
0.2
0.1
0.0
0.1
0.2
[CSR
(θsu
n=
0)
- C
SR
(θsu
n)]
/ C
SR
(θsu
n)
αcir =0.5
αcir =1.0
αcir =2.0
αcir =3.0
αcir =5.0
Figure 3.17: Relative deviation of CSR calculated at θsun = 0 and θsun 6= 0 but for thesame slant path optical thickness for HEY solid-columns. For detailed description see text.
The relative deviation of the CSR calculated from simulations at θsun = 0 to the reference
simulations with θsun 6= 0 are shown in Fig. 3.17 for cirrus clouds and in Fig. 3.18 for
aerosol. Considering the cirrus clouds, the deviations stay mostly confined to ±5%. The
largest calculated deviation amounts to -21% so that it can be concluded that deviations in
3.5 Parametrization of Circumsolar Radiation 65
0 10 20 30 40 50 60 70 80θsun /
0.1
0.0
0.1
0.2
0.3
0.4
0.5
[CSR
(θsu
n=
0)
- C
SR
(θsu
n)]
/ C
SR
(θsu
n)
αcir =0.5
αcir =1.0
αcir =2.0
αcir =3.0
αcir =5.0
Figure 3.18: Relative deviation of CSR calculated at θsun = 0 and θsun 6= 0 but for thesame slant path optical thickness for several aerosol components. For detailed descriptionsee text.
the order of 20% must be expected for individual cases. While for cirrus clouds the devia-
tions are distributed around the ±0, an overestimation of the CSR at high sun zenith angle
was diagnosed for most aerosol setups. To avoid introducing a bias in the results, k-values
for aerosol were additionally tabulated at θsun = 60, 70 and 80. In the computation of
CSR this requires an additional linear interpolation step in θsun.
3.5.4 Sensitivity of the CSR Parameterization on the Scattering
Layer’s Geometry
If not stated otherwise, cirrus clouds were always placed in a layer between 10 km and 11 km
above the ground in the MYSTIC simulations. Aerosol was always evenly distributed
in a layer of one kilometer depth directly above the ground. This was founded on the
assumption that circumsolar radiation is caused mainly due to single scattering and that
Rayleigh scattering contributes only negligibly to the signal. For example, Lohmann et al.
(2006) also reported that the cloud geometrical thickness has negligible influence on global
horizontal and direct irradiance at the surface. Here this is verified in regard to circumsolar
radiation for some scenes by exemplary calculations of the CSR (αcir = 3). Considering
clouds, the cloud height as well as the cloud thickness were varied in the control simulations.
66 3. Tools and Methods
For aerosol, only impact of a change of the thickness of the aerosol layer was assessed,
because most of the aerosol will normally be concentrated in the atmospheric boundary
layer and not in elevated layers.
Ice clouds composed of HEY solid columns were simulated in a layer 0 km and 1 km above
ground. The simulations were then compared to the original simulations. This was done
for 92 combinations of slant path optical thickness varying between 0.2 and 4.0, effective
radius varying between 5 µm and 90 µm and sun zenith angle varying between 0 and 78.
The deviations from the originally calculated CSR – i.e. for the cirrus in 10 km height –
were always below 2% or 0.003 which is below the calculated Monte Carlo uncertainty for
most simulations performed for this sensitivity study. Furthermore cirrus clouds of only
300 m thickness between 10.0 km and 10.3 km were simulated using the same parameter
combinations. The deviations in CSR from the original simulations with clouds of 1000 m
thickness were again below 2% or 0.003.
miam, ssam and sscm12 aerosol from OPAC was simulated in a 300 m thick layer above
ground. Following, the CSR (αcir = 3) was compared to the original simulation with a
1 km thick aerosol layer. In the simulations slant path optical thickness values between 0.1
and 3.0 were considered. The sun zenith angle was varied between 0 and 78. In total
simulations for 90 parameter combinations were compared. In all cases the simulations for
the thin aerosol layers deviate by less than 1% from the simulations for the thick aerosol
layer. These exemplary verification simulations confirm that the circumsolar radiation is
not unduly sensitive to the geometry of the scattering layer.
3.5.5 Further Aspects of the Apparent Optical Thickness Ap-
proach
Shiobara and Asano (1994) showed that k (Eq. 3.14) can be approximated by evaluating
the single scattering phase function P :
k ≈ 1− ω0
∫ αcir
0P (θ) sin(θ)dθ∫ π
0P (θ) sin(θ)dθ
(3.28)
with ω0 being the single scattering albedo. I verified this with the cirrus optical properties
considered in this study at one wavelength (550 nm) by performing MYSTIC simulations.
12miam: mineral dust accumulation mode. ssam: sea salt accumulation. sscm: sea salt coarse mode
3.6 Sensitivity of Circumsolar Radiation to Ice Particle Shape 67
The approximation leads to deviations of less than 5% in k(550 nm) as long as τs < 3 and
αcir > 0.5. For smaller angles the extent of the sun which is not captured by the ap-
proximation causes larger deviations from the values obtained from MYSTIC simulations.
Therefore the more exact results from MYSTIC were used to compute k in this study.
If we consider a non-absorbing medium (ω0 = 1), Eq. 3.28 can also be written as
k ≈∫ παcir
P (θ) sin(θ)dθ∫ π0P (θ) sin(θ)dθ
. (3.29)
This means that k is approximated by the relative amount of radiation that is scattered
out of the forward direction (i.e. the circumsolar region) by single scattering.
Interestingly the concept of the apparent optical thickness (Sect. 3.5.1) which was origi-
nally born out of shortcomings in measuring direct radiation shows parallels to a concept
developed due to shortcomings in simulating diffuse radiation – the well known δ-scaling
approach (e.g. Joseph et al., 1976). The basic concept of δ-scaling is that the forward peak
of the phase function is truncated which is accounted for by a reduction of the optical
thickness. I.e. forward scattered radiation is treated as direct radiation.
3.6 Sensitivity of Circumsolar Radiation to Ice Par-
ticle Shape
Of the three parameters determining k (Eq. 3.15) for a cirrus cloud – field of view, effective
radius and ice particle shape – the field of view is the only one which is easy to determine.
The effective radius can be retrieved from MSG with APICS but is subject to uncertainties,
particularly for optically very thin clouds. The ice particle shape cannot be retrieved with
passive remote sensing methods like APICS at all. The uncertainties due to this are
evaluated in Sect. 4.1, looking at the complete circumsolar retrieval chain – including the
cloud property retrieval. In this section the uncertainties which arise in the modelling
of circumsolar radiation are evaluated separately. In Fig. 3.19 possible CSR-values are
depicted as a function of the field of view for a variety of cirrus clouds, each composed of a
random mixture of particles. The left panel shows values for clouds with τs = 0.4, the right
one for τs = 2.0. From the scatter of the data points one can deduce which uncertainties
arise in the determination of CSR if either the information about one parameter – particle
68 3. Tools and Methods
shape – or about both parameters – particle shape and effective radius – are absent. The
possible CSR values for the whole range of particle shapes and effective radii contained in
the k-LUT (see Sect. 3.1) are displayed: If the optical thickness is the only information
available, the whole band composed of the different symbols must be considered. In this
case the determined values have a large uncertainty. If the effective radius is known, the
range of possible values narrows to the band filled by the corresponding symbol type. The
remaining uncertainty originates from differences in the optical properties of the ice particle
shapes. This is the uncertainty that is inherent to the method even for an otherwise perfect
retrieval of the cloud properties τs and reff .
For the slant path optical thickness range of 0 < τs ≤ 3 uncertainties are also depicted
directly in Figs. 3.20 and 3.21. The first shows the difference ∆Icir between the maximal
and minimal possible values of the circumsolar irradiance in W m−2 depending on the field
of view and the optical thickness τs for all ice optical properties considered in the k-LUT.
The latter shows the relative uncertainties in CSR δCSR for the same parameters, computed
as
δCSR =CSRmax − CSRmin
0.5 · (CSRmax + CSRmin). (3.30)
In both graphs solid lines stand for a fixed reff of 25 µm and dashed lines for an undefined
effective radius. Again it becomes apparent that knowledge about the effective radius
reduces the uncertainties considerably.
3.7 Sensitivity of CSR to Aerosol Particle Shape and
Relative Humidity
As mentioned in Sect. 3.3.2, libRadtran comes with optical properties for spherical and
non-spherical dust particles. The non-spherical particles are modelled as prolate spheroids
which exhibit a surface equal to their spherical counterparts. This enables us to evaluate the
impact of the aerosol particle shape on the CSR. In Fig. 3.22 CSR values for spherical and
aspherical particles are compared considering the three different OPAC dust components.
The results are shown only for τs = 0.5, but they vary only marginally with changes in τs.
The differences between spherical and aspherical particles are mostly below 5%, except for
the mineral dust nucleation mode type for which the differences can reach up to 14% for
3.7 Sensitivity of CSR to Aerosol Particle Shape and Relative Humidity 69
Figure 3.19: Possible CSR-values as a function of the opening angle αcir for a variety ofcirrus clouds, each composed of a random mixture of the five particle shapes and the twoparticle mixtures (described in Sect. 3.1). Left: τs = 0.4. Right: τs = 2.0. Differentsymbols denote different effective radii (labeled Reff, units µm).
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5slant path optical thickness
0
20
40
60
80
100
120
140
160
∆I c
ir /
Wm−
2
αcir =0.38
αcir =0.50
αcir =1.00
αcir =2.00
αcir =4.00
Figure 3.20: Uncertainty in circumsolar radiation for certain field of views (legend givesopening half angle in degrees). Solid lines: For reff = 25 µm and undefined ice particleshape. Dashed lines: For undefined reff and ice particle shape.
70 3. Tools and Methods
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5slant path optical thickness
0
20
40
60
80
100
120
140
δ CSR /
%
αcir =0.38
αcir =0.50
αcir =1.00
αcir =2.00
αcir =4.00
Figure 3.21: Relative uncertainty δCSR for certain field of views (legend gives opening halfangle in degrees). Solid lines: For reff = 25 µm and undefined ice particle shape. Dashedlines: For undefined reff and ice particle shape.
small fields of view (αcir < 1). This indicates that as long as spheroid particles of equal
surface are considered (as in Wiegner et al., 2011), the variation in CSR is small. However
for more complex shaped particles further sensitivity studies should be considered.
Besides the aerosol type, the modelled aerosol particle size in the IFS depends on the rela-
tive humidity. This was outlined in Sect. 3.3.2 where the hygroscopic growth of some aerosol
components was discussed. Figure 3.23 shows how the hygroscopic growth translates into
CSR changes. CSR(αcir=2.5) for the different hygroscopic ECMWF aerosol components
was calculated at varying relative humidity. Thereby always the same aerosol dry mass
area loading was used corresponding to τs,550nm = 0.5. It shows that for some aerosol types
changes of approximately a factor of 2 compared to the CSR values at 50% RH must be
expected due to changes in the relative humidity.
3.7 Sensitivity of CSR to Aerosol Particle Shape and Relative Humidity 71
0 1 2 3 4 5αcir /
0.98
1.00
1.02
1.04
1.06
1.08
1.10
1.12
1.14
CSR
spher
ical/C
SR
aspher
ical minm
miam
micm
Figure 3.22: Parameterized CSR for varying field of view obtained for spherical dust par-ticles divided by CSR for aspherical dust particles. Three curves for the different OPACdust aerosol components: Dotted blue: Mineral dust nucleation mode (minm), Solid green:Mineral dust accumulation mode (miam), dashed red: Mineral dust coarse mode (micm).
0 20 40 60 80 100relative humidity / %
0.5
1.0
1.5
2.0
CSR
/ C
SR
(rH
=5
0%
)
SS small
SS medium
SS large
OC
Sulfate
BC
Figure 3.23: Parameterized CSR at constant aerosol area dry mass loading for varying rela-tive humidity (RH) divided by CSR at 50% RH. Different lines for the different hygroscopicECMWF aerosol components. SS: sea salt, OC: organic carbon, BC: black carbon.
72 3. Tools and Methods
Chapter 4
Results
The results of the study are presented in this chapter. It starts with an intercomparison
of cirrus related CSR for different ice particle shapes (Sect. 4.1). Next, results for aerosol
and cirrus related CSR are validated against ground measurements performed at the PSA
(Sect. 4.2). Finally, in Sect. 4.3 some characteristics of cirrus related circumsolar radiation
are derived. For aerosol this step was omitted, as the validation proofed the results not to
be reliable enough (see also discussion in Sect. 5.1).
4.1 Intercomparison of Retrieval Results for Different
Ice Particle Shapes
In Sects. 3.1 and 3.6 it was outlined that uncertainties exist in the cloud property retrieval
as well as in the conversion of these properties to CSR values due to the unknown ice particle
shape (mixture). It is not obvious how these uncertainties interact, in particular, whether
they cancel at least partially. Therefore the overall variability was investigated by applying
the complete retrieval chain yielding circumsolar radiation from SEVIRI measurements
several times using different setups.
To this end CSR values for a limiting angle αcir = 2.5 have been retrieved from one year
(May 2011 – Apr. 2012) within the SEVIRI test sector (Fig. 2.4). This was performed
for the five ice particle shapes and the two shape mixtures described in Sect. 3.1. I.e.
seven distinct APICS runs were performed with different cloud optical properties. In the
following conversion from the retrieved cloud properties to CSR values, optical properties
74 4. Results
for the same particle shape were used as in the correspondent APICS run, so that seven
different CSR datasets were obtained in total.
A histogram of the occurrence of CSR values relative to all time steps with θsun < 80 was
computed for every ice particle shape, considering the whole domain. These histograms
are shown in the upper panel of Fig. 4.1 which gives a first illustration of the uncertainty
induced by not knowing the particle shape. The “Baum” optical properties are based
on a realistic mixture of particle shapes, so that an operational retrieval would rely on
them rather than on individual particle shapes. Therefore the corresponding lines are
highlighted. The lower panel of Fig. 4.1 shows basically the same histogram, but this time
only values contribute that go along with total irradiance values Itot,α=2.5 > 200 W/m2.
I considered this to be the lower operation limit of a hypothetical CST plant. Here the
occurrence is shown relative to all time steps that fulfil Itot,α=2.5 > 200 W/m2, i.e. inclusive
corresponding clear sky time steps. For the calculation of Itot,α=2.5 the clear sky direct
irradiance I0 is required (Eq. 3.19). It was obtained from libRadtran calculations which
were performed with a θsun-resolution of 1 and interpolated in between. Thereby the
elevation was set to 0 m above sea level. In the presence of water clouds, Itot,α=2.5 was
assumed to be below the 200 W/m2 threshold because water clouds are at most times
optically too thick to allow higher values. These cases were identified with a general cloud
mask from EUMETSAT (EUM OPS DOC 09 5164; EUM MET REP 07 0132). Pixels
marked cloudy in this mask but not in the COCS cirrus cloud mask were assumed to
contain water clouds. While on average 22% of the SEVIRI measurements in the test
data set produce a cirrus cloud detection, only 8%-10% additionally satisfy the 200 W/m2
criterion (depending on the assumed ice particle shape).
To gain insight into how much the retrieval results scatter, Fig. 4.2 shows, as example, the
distribution of CSR from the retrieval with “Baum v2.0”, which was applied to a subset
of SEVIRI measurements – namely to the subset for which the retrieval with “Baum
v3.5” yielded CSR values between 0.17 and 0.18. Ideally, if the ice particle shape had no
influence, all “Baum v2.0” results would also fall into the same interval. However in reality
the distribution is clearly wider.
Figure 4.3 gives a more complete comparison of the scatter between the different optical
properties: For this purpose, the SEVIRI measurements were binned into CSR intervals
(retrieved using “Baum v3.5”) of width 0.01. Each subset is then processed again assuming
other cloud optical properties (specified in the y-axis label of each panel). The resulting
new CSR distribution is then color-coded along the y-axis. Figure 4.2 shows the “Baum
4.1 Intercomparison of Retrieval Results for Different Ice Particle Shapes 75
Figure 4.1: Histograms of the relative occurrence of CSR values in case of cirrus cloudsfor one year within the test sector for different assumed ice particle shapes. Results forthe “Baum” ice particle mixtures are shown colored. Results for the different individualice particle shapes from HEY are shown in grey. Upper panel: Histograms of the relativeoccurrence with respect to the total number of SEVIRI measurements with θsun < 80.Lower panel: Histograms of the relative occurrence of CSR values under the conditionItot,α=2.5 > 200W/m2 (lower operation limit of the assumed hypothetical CST plant) withrespect to the total number of SEVIRI measurements with Itot,α=2.5 > 200W/m2.
76 4. Results
Figure 4.2: Normalized distribution of retrieval results using “Baum v2.0” for SEVIRImeasurements for which the retrieval with “Baum v3.5” yielded values between 0.17 an0.18. Dashed lines mark the 25th and 75th percentiles (q25 & q75).
v2”-CSR-distribution within one of these subsets and is simply a vertical cross section
through the upper left panel. From Fig. 4.3 one can see that droxtals yield quite different
CSR results than “Baum v3.5”: The results within the individual CSR subsets scatter
widely and most times droxtals yield smaller CSR values than “Baum v3.5” (distribution
is curved to lower values). Hollow columns and rough aggregates have rather narrow
distributions but also show some curvature – implying a bias compared to “Baum v3.5”.
Rosettes produce a relatively narrow distribution close to the 1:1 line. The distributions
for “Baum v2.0” and solid columns are quite similar, being rather broad with no clear bias
visible.
4.2 Validation
In the following parameterized CSR values (CSRparam) are compared to ground measure-
ments of the CSR (CSRSAMS) performed at the Plataforma Solar de Almerıa with SAMS
(Sect. 2.7). First the CSR parameterized from ECMWF IFS modelled aerosol data is eval-
uated in Sect. 4.2.1, before the satellite based CSR parameterization for clouds is reviewed
in Sect. 4.2.2. The SAMS measurements show varying frequency but were generally per-
formed at least once per minute, except for a few days which had to be discarded due to
4.2 Validation 77
Figure 4.3: Normalized CSR retrieval distributions for 100 bins of CSR from “Baum v3.5”(x-axis, bin width 0.01). Dashed lines mark q25 and q75, i.e. they enclose 50% of themeasurements. Black vertical line in upper left panel marks the cross section that isdisplayed in Fig. 4.2.
78 4. Results
technical failures of the instrument. Unless stated otherwise, the CSR refers to an opening
half-angle of 2.5 in this section and one year of data was evaluated (May 2011 – Apr.
2012).
Following measures were used for an objective validation: The relative bias, the mean ab-
solute deviation (MAD), the root-mean-square deviation (RMSD), the Pearson correlation
r (Wilks, 2005, Eq. 3.23) and the Spearman rank correlation rrank (Wilks, 2005, Eq. 3.28).
The RMSD and the Pearson correlation are provided since they are widely used validation
measures, however it should be noted that they have to be interpreted with care: The
deviations between the SAMS measured and the satellite retrieved values are not necessar-
ily normally distributed such that an interpretation of the RMSD in terms of a standard
deviation may be misleading. Furthermore the RMSD strongly penalizes outliers which,
for example, can easily be caused by collocation errors of satellite data. Also can a truly
linear relationship between the CSR from SAMS and from the developed parameterization
not be expected which may deteriorate Pearson correlation values. This is because errors
in ∆k or τs will not propagate linearly (comp. Eq. (3.22)) – at least not for cirrus clouds
with ∆kατs being not much smaller than 1. For N evaluated CSR tuples the corresponding
formulas are:
Bias =
∑Ni=1 CSRparam,i −
∑Ni=1 CSRSAMS,i∑N
i=1 CSRSAMS,i
(4.1)
MAD =N∑i=1
|CSRparam,i − CSRSAMS,i |N
(4.2)
RMSD =
√√√√ N∑i=1
(CSRparam,i − CSRSAMS,i)2
N(4.3)
r =
1
N − 1
∑Ni=1
[CSR′param,iCSR′SAMS,i
][
1
N − 1
∑Ni=1
(CSR′param,i
)2]1/2 [
1
N − 1
∑Ni=1
(CSR′SAMS,i
)2]1/2
, (4.4)
where primes denote anomalies from the mean values. The Spearman rank correlation rrank
is simply the Pearson correlation applied not on the data but on the ranks of the data. It is
not a measure of linear relationship but of monotone relationship. The Pearson correlation
is neither robust nor resistant while the Spearman rank correlation is – contrasting the
Pearson correlation it requires no assumption about the kind of monotone relationship and
4.2 Validation 79
is not unduly influenced by few outliers (Wilks, 2005, Chap. 3).
4.2.1 Validation of Aerosol Related Circumsolar Radiation
In this section, CSR parameterized from aerosol data from the IFS near-real-time forecast is
compared to SAMS measurements. No information about the relative humidity is included
in the NRT data. Therefore the parameterized values refer to a relative humidity of 50%,
unless stated otherwise.
The cloud free SAMS measurements, falling in the timespan of ±7.5 min around the quarter
hour, were averaged by calculating their arithmetic mean. This reduces potential random
measurement errors, and allows for a comparison to the parameterized CSR values, which
were also computed by the quarter hour (comp. Sect. 3.3.3). The cloud free state was
initially assessed via the cloud flag in the SAMS dataset. IFS output was taken from the
“nearest neighbour” column of the model to the PSA. Evaluating one year (May 2011 –
Apr. 2012), 7972 data tuples were generated.
If the parameterization is applied straight on, the resulting CSR values are on average
clearly too low. The bias is -73%. In the histogram shown in Fig. 4.4 the parameterized
CSR values stay mostly confined to the first two bins (below 0.01) while the measured
CSR is broader distributed. Not only are the parameterized CSR values too low, they
also correlate only weakly with the measured values. For the initially processed 7972
data tuples the Spearman rank correlation is only 0.25. It turns out, that the measured
CSR often fluctuates considerably on the scale of minutes, even if no cloud is detected by
SAMS. Examination of sky images from an automated camera positioned besides the SAM
instrument revealed one possible reason for this. The CSR fluctuations are often caused by
non-condensing convection, which appears as highly variable haze plumes, but sometimes
also by clouds that are not detected by SAMS. These fluctuations occur on scales that are
not resolved by the model. Furthermore both phenomena can cause increased CSR values
which cannot be captured by the parameterization.
To evaluate how the parameterized CSR performs in a stable, purely aerosol dominated
situation the data set is filtered: If there are measured CSR values within the 15 min
averaging period which deviate by more than 50% from the mean for this period, this
period is discarded from evaluation. Furthermore, if the mean CSR for a period deviates
too much from the preceding or following one, it is discarded as well: Only periods are
80 4. Results
0.00 0.02 0.04 0.06 0.08 0.10CSR(αcir =2.5 )
0
1000
2000
3000
4000
5000
6000
#
ECMWF IFS
SAMS Measurement
Figure 4.4: Blue: Histogram of CSR(αcir = 2.5) for the PSA due to aerosol, parameterizedfrom IFS output. Green: Histogram of CSR(αcir = 2.5) measured by SAMS at the PSAat times without a cloud flag in the SAMS dataset, averaged over 15 min. 7972 data pointsin total. 253 data points of SAMS CSR > 0.10 not shown.
considered for which the relative change in CSR · cos(θsun) is smaller than 5%. Here, the
CSR is multiplied by cos(θsun) to equalize deviations caused solely by a changing sun zenith
angle to first order. To ensure stable clear sky conditions, it is further required that at least
four averaging periods in a row meet these conditions. Data tuples belonging to smaller
contiguous groups are discarded as well. This stringent filtering leaves 1081 data points.
Still, the parameterized values are on average too low (Bias -67%). Also the Spearman
correlation is quite low (30%).
One reason for the poor agreement between the datasets lies probably in the underesti-
mation of the concentration of the largest aerosol particles by the IFS. The largest aerosol
particles are most relevant for the forward scattering peak but not necessarily for the
optical thickness. Indeed it is known that the evaluated model version suffers from too
much small dust particles at the expense of the larger ones (pers. comm. J.-J. Morcrette).
Nevertheless, a comparison of the Angstrom exponent α shows that the model can dis-
4.2 Validation 81
criminate between situations with large or small aerosol particles dominating scattering.
The Angstrom exponent is a qualitative indicator of particle size. It is defined as
α = −log
τλ1τλ2
logλ1
λ2
, (4.5)
with τλ1 and τλ2 being the aerosol optical thickness at the different wavelengths λ1 and λ2.
Angstrom exponents / 1 indicate that the size distribution is dominated by large particles
with radii ' 0.5 µm, or vice versa (Schuster et al., 2006). The Angstrom exponent was
calculated from IFS aerosol area mass loadings utilizing the aerosol optical properties
created for this study. In Fig. 4.5 it is compared to the Angstrom exponent computed
from aerosol optical thickness values determined by the AERONET-processing scheme1
(Holben et al., 1998), which is applied to the Cimel sun photometer data from the PSA.
The IFS is able to coarsely reproduce the temporal evolution of the Angstrom exponent
measured at the PSA, although it exhibits on average lower values. At first this may seem
contradictory to the CSR bias, as a smaller Angstrom exponent in principle means larger
particles. However, it should be mentioned that while the Angstrom exponent tells if coarse
mode aerosol is present or not, it cannot be used to determine whether the concentration
of the largest particles is indeed correct. This is because the scattering efficiency – and
therefore the Angstrom exponent – exhibits only a weak dependence on particle size for
particles considerably larger than the wavelength of the light.
One has to realize that the parameterization of CSR from IFS modeled aerosol data cannot
be applied straight on. Two reasons have been identified for this: On one hand the CSR
often fluctuates substantially on scales that are not resolved by the model, even in clear sky
conditions. On the other hand, the parameterized values are considerably too low, even for
stable clear sky conditions. Also the disregard of the hygroscopic growth of some aerosol
components cannot explain the strong bias: The difference in mean CSR parameterized
assuming a constant relative humidity of 50% and 99% amounts to about 25%.
For these reasons, the IFS dataset was not evaluated globally because the results would
probably be not reliable. However, in Sect. 5.1 the possibility of a statistical post-processing
of the IFS output to obtain reasonable CSR data is discussed.
1Level 1.5 data, i.e. automatically cloud screened, but final a posteriori calibration not applied.
82 4. Results
May 06 2011
May 13 2011
May 20 2011
May 27 2011
Jun 03 2011
Jun 10 2011
Jun 17 2011
Jun 24 2011
Jul 01 20110.0
0.5
1.0
1.5
2.0
Angst
roem
exponent
IFS
AERONET
Figure 4.5: Green dots: Angstrom exponent (440 nm/870 nm) calculated from aerosoloptical thickness values measured by the AERONET Cimel sun photometer at the PSA(Level 1.5 data). Blue dots: Angstrom exponent (440 nm/870 nm) calculated from IFSaerosol concentration for the corresponding model column utilizing the optical propertiesdescribed in Sect. 3.3.2.
4.2.2 Validation of Cirrus Related Circumsolar Radiation
In the following CSR retrieved from SEVIRI measurements for cirrus covered sky are
compared to ground measurements of the CSR performed with SAMS at the PSA.
The comparison of satellite data with ground measurements is not trivial due to the dif-
ferent spatial footprints and viewing geometries of the instruments. To get a best possible
match methods discussed by Greuell and Roebeling (2009) were applied: A correction of
the parallax displacement, which results from the two instruments (SAMS & SEVIRI) look-
ing at an elevated cloud layer under different geometries, was used2. A schematic drawing
of the measurement geometry of SAMS and SEVIRI in Fig. 4.6 illustrates this parallax
correction of the SEVIRI measurements. In case of circumsolar radiation measurements a
parallax correction must consider not only the satellite viewing geometry but also the sun
geometry since SAMS looks into the sun. For these corrections the cirrus was assumed
to lie between altitudes of 9 and 11 km. First, the algorithm determines the geographic
location of the atmospheric volumes between 9 and 11 km corresponding to the SEVIRI
2Courtesy of Luca Bugliaro, DLR
4.2 Validation 83
pixels which are geo-referenced at 0 km altitude. This step depends on the viewing geo-
metry of SEVIRI. In the following, those SEVIRI pixels are selected whose corresponding
elevated atmospheric volumes are crossed by the line of sight between SAMS and the sun.
If more than one SEVIRI pixel is identified as relevant, the retrieved optical thickness and
effective radius values of the pixels are averaged. However, the effective radius is averaged
only over cirrus covered pixels. To reduce short term variability which can deteriorate the
validation results if the SEVIRI and SAMS measurements are not perfectly collocated, the
already parallax corrected cloud property retrieval results from three consecutive SEVIRI
measurements at times t0 − 15 min, t0 and t0 + 15 min were temporally averaged, before
the CSR was computed from this average. Collocation errors can be caused by wrong as-
sumptions about the cloud height and thickness but also by deviations in the registration
of SEVIRI pixels to latitude/longitude values (comp. Sect. 2.3.1).
The SAMS measures CSR only at one place while the method developed in this study
delivers an average value for several square kilometers. To alleviate this scale discrepancy
the ground measurements were brought to the same time grid as the SEVIRI measurements
by averaging them within a symmetrical time span ∆t around t0. The averaging of the CSR
from SAMS was performed applying a Gaussian weighting function w to the measurements
w = exp
[−2(t− t0)2
∆t2
](4.6)
with t being the time of the SAMS measurement and t0 being the time of the central
SEVIRI measurement. Underlying this is the assumption that a temporal average of the
advected cloud properties is representative of the spatial average of cloud properties re-
trieved at distinct times from SEVIRI. I found the averaging time of 35 min to deliver
better agreement between satellite and ground measurements than without the averaging,
and did not see considerable improvement for longer averaging times. Therefore I used this
value for ∆t. Only SAMS measurements meeting the condition
|t− t0| < ∆t = 35 min (4.7)
were considered. The Gaussian therefore has not to be evaluated to ±infinity, but is cut
off at these limits.
Before comparing CSR values for a 2.5 field of view the measurements were filtered: Only
time steps were considered with a positive cirrus cloud detection from MSG. I.e. at least
one of the SEVIRI pixels considered in the parallax correction process at t0 − 15 min, t0
84 4. Results
9km
2km
A
FE
CB
D
Figure 4.6: Schematic drawing of the different measurement geometries of SAMS andSEVIRI (not to scale) as illustration of the parallax correction of SEVIRI measurements.A: SAMS. B: MSG satellite. C: Sun. D: Original SEVIRI grid referenced on the ground.E and F: Atmospheric volumes between 9 km and 11 km probed by SEVIRI and SAMS. Ifa cirrus is detected from MSG within E but not within F, as illustrated in the drawing, theretrieved optical thickness is averaged over E & F. The effective radius is solely averagedover cloudy pixels, for the illustrated case this means that the value of E is preserved.
4.2 Validation 85
and t0 + 15 min was required to be cloudy. Furthermore I required the total irradiance
Itot,α=2.5 calculated from the averaged SEVIRI measurements to be above 200 W/m2 to
ensure relevance for CST plants.
Excerpts of the filtered time series for three days are shown as example in Fig. 4.7. From
the figure it becomes apparent that the temporal evolution of the CSR measured by SAMS
is in general captured by the satellite time series, but there are timing and amplitude errors
so that the CSR values at a given time often disagree.
04:00:00
06:00:00
08:00:00
10:00:00
12:00:00
14:00:00
16:00:00
18:00:00
20:00:00
22:00:000.0
0.1
0.2
0.3
0.4
0.5
CSR
2011-06-05
SAMS
APICS Baum 3.5
04:00:00
06:00:00
08:00:00
10:00:00
12:00:00
14:00:00
16:00:00
18:00:00
20:00:00
22:00:00
2011-06-06
04:00:00
06:00:00
08:00:00
10:00:00
12:00:00
14:00:00
16:00:00
18:00:00
20:00:00
22:00:00
2011-06-20
Figure 4.7: Exemplary excerpt of the CSR time series used for the validation. Time isgiven as UTC.
In Tab. 4.1 the different validation measures are listed which were introduced at the begin-
ning of this chapter. The rank correlation between 0.48 and 0.54 (depending on employed
optical properties) and the MAD of 0.11 show that there are often considerable differences
between the time series if instantaneous values at a given time are compared. However,
the statistical values of CSR are more important than the instantaneous ones for long-term
CST system performance prediction (e.g. Rabl and Bendt, 1982). Therefore, Fig. 4.8 com-
pares the histograms of CSR from SAM and retrieved from SEVIRI using “Baum v3.5”.
Although there are some differences, the histograms compare well considering the two
completely different methods operating on different scales. It is also encouraging that the
bias value for both employed ”Baum” optical property sets is close to zero (4% and -11%
respectively, see Tab. 4.1). With Figs. 4.7 & 4.8 in mind, I therefore conclude that the
presented method is suitable to generate typical distributions or time series of the cirrus
related CSR.
To locate reasons for the disagreement between the two time series considering instanta-
86 4. Results
neous values, sky images from an automated camera positioned besides the SAM instru-
ment were examined. After a manual inspection of the data series and the sky images it
was suspected that presence of water clouds (in most cases cumuli) might compromise the
results. The cumulus clouds – also if only covering a SEVIRI pixel partly – will increase
the reflectivity in the SEVIRI 0.6 µm channel. This will result in APICS retrieving an in-
creased optical thickness for the cirrus. This hypothesis was tested with a shortened version
of the evaluation time series (01 May 2011 – 30 Jun. 2011). For this period the data were
additionally filtered by hand, leaving only slots without cumuli visible in the sky camera
images. Table. 4.2 contains the validation measures for the shortened time series. The
numbers within parentheses in Tab. 4.2 were obtained without the manual cumulus clouds
screening as reference. The validation measures are improved by the manual filtering: The
rank correlation and correlation are increased and the MAD is decreased compared to the
unfiltered dataset. Figures 4.9 and 4.10 show two examples from the series of sky images
which was manually evaluated for the cumulus cloud screening. On 06 Jun. 2011 (Fig. 4.9)
clouds at different atmospheric levels with high spatial variability were present during most
of the day. Consequently this day was removed from the time series in the manual cumulus
cloud screening. On this day pronounced disagreements between the two time series are
observable (comp. Fig. 4.7). In the image in Fig. 4.10 the sky appears rather homogeneous.
The clouds are clearly identified as cirrus clouds by the presence of a halo. The image was
acquired at 08 UTC on 20 Jun 2011. This date is also depicted in Fig. 4.7. At this time
the parameterized CSR matches the measured one. This limited evaluation indicates that
an automated cloud detection excluding pixels contaminated with water clouds would be
beneficial for future applications of the circumsolar radiation retrieval.
Table 4.1: Results of the comparison of CSR measured from ground and retrieved fromSEVIRI for different setups: Rank correlation rrank,CSR, Pearson correlation rCSR, meanabsolute deviation MAD, root mean square deviation RMSD, bias and the number ofcompared data tuples N .
Optical Properties rrank,CSR rCSR MAD RMSD Bias NBaum v2.0 0.54 0.50 0.11 0.16 4% 2021Baum v3.5 0.48 0.44 0.11 0.15 -11% 1890
4.2 Validation 87
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8CSR(αcir =2.5 )
0
100
200
300
400
500
600
700
800
# N = 1890
APICS Baum 3.5
SAM Measurement
Figure 4.8: Histograms of CSR measured by SAMS (green) and retrieved from SEVIRI(blue) as used for the validation in Sect. 4.2.2.
Figure 4.9: Image of the sky acquired by an automated camera at the PSA on 06 Jun2011 at 12 UTC (Courtesy of Stefan Wilbert, DLR).
88 4. Results
Figure 4.10: Image of the sky acquired by an automated camera at the PSA on 20 Jun2011 at 08 UTC (Courtesy of Stefan Wilbert, DLR).
Table 4.2: Same as Tab. 4.1 but for the shortened time series (01 May 2011 – 30 Jun.2011) which was manually cumulus screened. Values in parenthesis are for unscreenedtime series.
Optical Properties rrank,CSR rCSR MAD RMSD Bias NBaum v2.0 0.79 (0.68) 0.75 (0.62) 0.08 (0.10) 0.12 (0.14) 18% (14%) 220 (407)Baum v3.5 0.76 (0.65) 0.67 (0.58) 0.07 (0.09) 0.10 (0.13) -7% (-7%) 213 (386)
4.3 Characteristics of Cirrus-Related Circumsolar Radiation 89
4.3 Characteristics of Cirrus-Related Circumsolar Ra-
diation
4.3.1 Spatial Distribution
The average circumsolar irradiance and CSR during CST plant operation is strongly de-
pendent on the lower DNI limit the plant can work at. To provide an example of the
spatial distribution of the circumsolar radiation, I consider again a hypothetical power
plant with a lower operation limit of Itot,α=2.5 > 200 W/m2 like in Sect. 4.1. Figure 4.11
shows two maps of the circumsolar irradiance Icir,α=2.5 averaged over all time steps with
Itot,α=2.5 > 200 W/m2 – one for the “Baum v3.5” optical properties and one for “Baum
v2.0”. The values for “Baum v2.0” are on average about 50% higher than for “Baum
v3.5”, but the regional distribution patterns are similar. Therefore intra-site comparisons
are possible with considerably less dependence on the assumed ice particle shape. The
shown maps highlight a distinct small scale variability, i.e. on the scale of few SEVIRI pix-
els. This variability cannot be resolved by ground measurements at few selected places. In
the presented example the “Baum” parameterizations mark the extremes of the temporally
and spatially averaged values and the HEY parameterizations lie in between (not shown
here). There are few red pixels visible in the figure standing out from their environment.
These are caused by unusually frequent cloud detections by COCS at these locations (see
also Sect. 5.2.2).
Figure 4.12 emphasises that the lower DNI operation limit of a CST system has indeed
strong influence on the average circumsolar radiation during operation. Both panels in the
figure show the average cirrus related CSR for αcir = 2.5 and “Baum v3.5” during opera-
tion of a hypothetical power plant. However, they differ in the assumed operation limits:
While in the left panel CSR is averaged over all time steps with Itot,α=2.5 > 200 W/m2, the
CSR shown in the right panel is an average over all time steps with Itot,α=2.5 > 100 W/m2.
The spatial patterns are similar, but the magnitude of the values for the 100 W/m2-
threshold are considerably higher. The mean values averaged over the whole domain are
0.0164 and 0.0272, respectively. I.e. the change of the operation limit from 200 W/m2 to
100 W/m2 causes an average increase of 66% in the CSR. A change of the opening angle
αcir has less effect in this case: If we consider the same time steps as in the left panel
(Itot,α=2.5 > 200 W/m2) but calculate the CSR for αcir = 0.7 and αcir = 5.0 the values
90 4. Results
0 50 100 150pixel
0
50
100
150
200
250
pix
el
Baum 2.0
0
2
4
6
8
10
12
14
16
18
20
I cir,α
=2.
5 /
Wm−
2
0 50 100 150pixel
0
50
100
150
200
250
pix
el
Baum 3.5
0
2
4
6
8
10
12
14
16
18
20
I cir,α
=2.
5 /
Wm−
2
Figure 4.11: Circumsolar irradiance for a limiting angle of 2.5, averaged over all time stepsin the test dataset with Itot,α=2.5 > 200 W/m2 obtained using the “Baum v3.5” (left) and“Baum v2” (right) optical property datasets.
change moderately to 0.0113 and 0.0183, respectively. This corresponds to a change of
-33% and +12% in regard to the value for αcir = 2.5 (compare also Fig. 3.19).
In this study evaluation of SEVIRI data was limited to a one year period for the sake
of reduced computing time and storage space. Processing of longer time spans was not
performed because the results would be most likely not directly applicable in a specific solar
resource assessment process, which depends on the CST facility design. The presented
results serve primarily as a proof of concept and may deviate from the long term mean.
Still they allow to estimate the magnitude of the considered effects and give insight on the
importance of the regional distribution. To obtain a rough estimate for the inter-annual
variability of circumsolar irradiance due to cirrus clouds one can evaluate the variability
of the cirrus cloud occurrence itself. To this end, I calculated the mean cirrus cloud cover
within the test sector from the COCS cloud mask for six time spans of a full year’s length
available (years 2006 to 2010 and the time span from May 2011 to Apr. 2012). The
mean annual anomaly of the cirrus cloud cover derived from this data lies between 8%
and 28%, depending on location, with a spatial average of 13%. The cirrus cloud cover for
the time span May 2011 to Apr. 2012, for which circumsolar radiation was calculated, is
on average 11% lower than the mean value for the six considered time periods. One can
also contemplate the DNI variability: Lohmann et al. (2006) derived DNI from data of the
4.3 Characteristics of Cirrus-Related Circumsolar Radiation 91
0 50 100 150pixel
0
50
100
150
200
250
pix
el
Icir,α=2.5 >200W/m2
0.000
0.006
0.012
0.018
0.024
0.030
0.036
0.042
0.048
0.054
0.060
CSR
(αci
r=
2.5)
0 50 100 150pixel
0
50
100
150
200
250
pix
el
Icir,α=2.5 >100W/m2
0.000
0.006
0.012
0.018
0.024
0.030
0.036
0.042
0.048
0.054
0.060
CSR
(αci
r=
2.5)
Figure 4.12: CSR for a limiting angle of 2.5, averaged over all time steps in the test datasetwith Itot,α=2.5 > 200 W/m2 (left) and Itot,α=2.5 > 100 W/m2, respectively, obtained usingthe “Baum v3.5” optical properties.
International Satellite Cloud Climatology Project (ISCCP). They reported annual standard
deviations of DNI for selected places between between 5% and 13%. From measurements
at the PSA Pozo-Vazquez et al. (2011) calculated an annual standard deviation of 6%
for DNI and 9% for the cloud fraction. With these numbers in mind, a uncertainty in the
range of 15% in the presented circumsolar radiation values in regard to the long-term mean
through natural variability seems to be a reasonable first-order estimate.
4.3.2 Irradiance Overestimation by Pyrheliometers
In this section it is characterised how much the irradiation usable by a CST system deviates
from the from pyrheliometer readings due to the cirrus related circumsolar radiation. At
first, a coarse proxy for the operating hours of a hypothetical power plant was calculated,
to estimate how much the solar resource is overestimated if irradiance is measured with a
pyrheliometer exhibiting a larger aperture than a CST plant. The mean number of time
steps at which Itot,αcir> 200 W/m2 – i.e. at which the power plant could be operating –
was computed for the test sector (Fig. 2.4) for αcir = 2.5 to represent a pyrheliometer
and αcir = 0.7 to represent a power plant. Depending on the assumed ice particle shape
this count is between 0.8% and 1.1% higher for the pyrheliometer compared to the power
92 4. Results
plant.
Furthermore the mean overestimation of usable irradiation itself by a typical pyrheliometer
due to cirrus clouds was calculated. To this end the irradiance integrated over calculated
operating time E was compared. Consider the integrals
E0.7 =
∫t0.7
Itot,αcir=0.7dt , (4.8)
E2.5 =
∫t2.5
Itot,αcir=2.5dt , (4.9)
where t0.7 and t2.5 are all time steps with Itot,αcir=0.7 > 200 W/m2 and Itot,αcir=2.5 >
200 W/m2, respectively. The relative overestimation of usable irradiance is then calculated
as
∆E =E2.5 − E0.7
E0.7(4.10)
The mean values of ∆E averaged over all evaluated pixels within the test sector lie between
0.4% and 0.7%, depending on assumed cirrus optical properties. Figure 4.13 shows his-
tograms of the (spatial) distribution of ∆E within the test sector (Fig. 2.4) for the optical
property datasets “Baum v2” and “Baum v3.5”. From these one can assess that on one
hand there can be considerable differences depending on location. E.g. considering the
histogram for “Baum v2”, one can see that there are pixels for which the overestimation
is calculated as low as 0.2%, but there are also others for which one must expect more
than 1%. On the other hand, the values seem approximately normally distributed so that
for most pixels the values cluster around a central value. To evaluate the sensitivity of
the resource overestimation, ∆E was also calculated for two additional hypothetical CST
systems. The first exhibits the same lower operation limit of 200 W/m2 but a smaller
acceptance angle αcir = 0.5. The second has the same acceptance angle of αcir = 0.7
but a changed lower operation limit of 100 W/m2. The average resource overestimation
calculated with “Baum v2” is increased by 34% and 3%, respectively, due to these config-
uration changes in regard to the original value with αcir = 0.7 and a lower operation limit
of 200 W/m2. In the presented example, the overestimation is clearly more sensitive to the
CST system’s aperture size than to the lower operation limit. In the previous Sect. 4.3.1,
however, average CSR values were calculated which proofed to be sensitive to the lower op-
eration limit. This highlights that a determination of which parameter is more important
4.3 Characteristics of Cirrus-Related Circumsolar Radiation 93
cannot be made with universal validity.
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6∆E / %
0
2000
4000
6000
8000
10000
12000
#Baum 2.0
Baum 3.5
Figure 4.13: Histograms of the relative overestimation of usable irradiation ∆E due tocirrus clouds by a pyrheliometer with αcir = 2.5 with regard to a hypothetical CSTsystem with αcir = 0.7 and a lower operation limit of Itot,αcir=0.7 > 200 W/m2. Blue:Values obtained utilizing the “Baum v2” cirrus optical properties. Green: Values obtainedutilizing the “Baum v3.5” cirrus optical properties.
Altogether, ∆E values mostly below 1% are small compared to other uncertainties in the
resource assessment process. For example, the long term mean value of the DNI could
be deduced only with an uncertainty of approximately 3% from a 10-year-long series of
perfect measurements, if one assumes an annual standard deviation of the DNI in the order
of 10% (postulating a normal distribution of the annual DNI values). Standard deviations
in this order are in line with reported values from Lohmann et al. (2006). In reality one will
usually have to cope with shorter time series and/or additional measurement errors so that,
depending on undertaken effort, deviations in the assessed DNI from the long-term average
in the order of 5% can occur (pers. comm. Richard Meyer, 2013). However, it should be
noted that the irradiance overestimation discussed in this section is no random uncertainty
but a systematic error which should be accounted for. Without delving into economics, one
should be aware of that the profit margin of a solar power plant can be strongly influenced
by variations in power generation. This is because the expenses for construction, operation
and maintenance are virtually independent from the produced power. Furthermore, the
price per unit energy is often pre-determined by long-term power purchase agreements or
feed-in tariffs (Meyer, 2013). The profit is usually small compared to the turnover and
can basically be regarded as the turnover minus the expenses. Therefore any influence on
94 4. Results
produced power has a leveraged effect on the profit. Following an estimate of Meyer et al.
(2008) even a 0.5% bias in the usable irradiance may lead to profit differences between
100 000e and 300 000e p.a. considering a typical Spanish solar thermal power plant
with 50 MWe∗. In general, the consideration of the resource overestimation becomes the
more important the lower the overall uncertainty can be fixed.
4.3.3 Relation of Irradiance and CSR
If typical time series of CSR shall be generated for the simulation of CST plants, its
frequency of occurrence must be linked to irradiance values. As there is no distinct depen-
dence between irradiance and CSR, the statistics of both measures should be considered
together. Figure 4.14 shows 2-dimensional histograms of irradiance and CSR obtained from
the SEVIRI test data set (only measurements considered with a positive cirrus detection
from COCS) and compares them to a histogram obtained from SAMS measurements (only
measurements considered with a positive cloud detection by SAMS). The histograms need
not necessarily be identical as the histograms for SEVIRI were computed for the whole
test domain while the SAMS measurements were performed only at the PSA. Furthermore
the cloud detection of SAMS is based basically on a threshold for the temporal variability
of the measured optical thickness and therefore it is unlikely that all detections are due to
cirrus clouds alone. The SAMS histogram exhibits a bifurcation for decreasing irradiance.
The lower part cannot be accounted to cirrus clouds with optical properties as used in
this study and is presumably caused by the aerosol contribution within falsely classified
clear sky measurements. The histogram obtained for “Baum v3.5” matches the SAMS
histogram better than the one for “Baum v2”. The latter shows a tendency to larger CSR
values compared to “Baum v3.5”.
In Fig 4.15 similar histograms are depicted for three different fields of view. They highlight
once more the dependence of circumsolar radiation of the field of view. Furthermore, the
DNI-CSR-combinations for cirrus clouds simulated by Thomalla et al. (1983) are given in
these plots to allow for a comparison. While in general they are within the range of values
retrieved from the SEVIRI data, there is a notable difference for the field of view with
αcir = 0.5. The high CSR values at around 800 W/m2 irradiance cannot be reproduced
with the cirrus optical properties considered in this study (Sect. 3.1).
∗Capacity of 50 MW electric power generation
4.3 Characteristics of Cirrus-Related Circumsolar Radiation 95
200 400 600 800 1000I tot ,α = 2.5 /iWm− 2
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
CS
R(α
cir=
2.5
)
10-5
10-4
10-3
10-2
rela
tiv
eio
ccu
rre
nce
200 400 600 800 1000I tot ,α = 2.5 /iWm− 2
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
CS
R(α
cir=
2.5
)
10-5
10-4
10-3
10-2
rela
tiv
eio
ccu
rre
nce
200 400 600 800 1000I CHP1/ CH1 /iWm− 2
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
CS
R(α
cir=
2.5
)
10-5
10-4
10-3
10-2
rela
tiv
eio
ccu
rre
nce
APICSiBaumiv2
APICSiBaumiv3.5
SAMS
Figure 4.14: 2-dimensional histogram of the occurrence of Itot,αcir=2.5 and CSR(αcir =2.5) in case of cirrus clouds relative to the amount of all cirrus cloud measurements.Upper panel: Retrieved using “Baum v2” optical properties. Middle panel: Retrievedusing “Baum v3.5” optical properties. Lower panel: Histogram from one year SAMSmeasurements (CloudPresenceVec=1). DNI measured with Kipp & Zonen CHP1/CH1.
96 4. Results
200 400 600 800 1000Itot,α=0.5 / Wm−2
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
CSR
(αci
r=
0.5)
10-5
10-4
10-3
10-2
rela
tive o
ccurr
ence
200 400 600 800 1000Itot,α=0.5 / Wm−2
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
CSR
(αci
r=
0.5)
10-5
10-4
10-3
10-2
rela
tive o
ccurr
ence
200 400 600 800 1000Itot,α=1.0 / Wm−2
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
CSR
(αci
r=
1.0)
10-5
10-4
10-3
10-2
rela
tive o
ccurr
ence
200 400 600 800 1000Itot,α=1.0 / Wm−2
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8C
SR
(αci
r=
1.0)
10-5
10-4
10-3
10-2
rela
tive o
ccurr
ence
200 400 600 800 1000Itot,α=5.0 / Wm−2
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
CSR
(αci
r=
5.0)
10-5
10-4
10-3
10-2
rela
tive o
ccurr
ence
200 400 600 800 1000Itot,α=5.0 / Wm−2
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
CSR
(αci
r=
5.0)
10-5
10-4
10-3
10-2
rela
tive o
ccurr
ence
Figure 4.15: 2-dimensional histogram of the occurrence of Itot,αcirand CSR(αcir in case of
cirrus clouds relative to the amount of all cirrus cloud measurements. Green dots: Valuesfrom Thomalla et al. (1983, Tab. 2b) (simulations without aerosol). Left panels: “Baumv2”, Right panels: “Baum v3.5”, Upper panels: αcir = 0.5, Middle panels: αcir = 1.0,Lower panels: αcir = 5.0.
Chapter 5
Discussion
In this chapter the possibility of a statistical post-processing of IFS aerosol data is discussed
(Sect. 5.1) as well as uncertainties in the cirrus cloud property retrieval (Sect. 5.2). In the
last section the cirrus related CSR results obtained from MSG measurements are compared
to a CSR parameterization from Neumann et al. (1998), which is based on DNI and GHI
measurements.
5.1 Statistical Post-Processing of IFS Aerosol Output
In Sect. 4.2.1 it was pointed out that the aerosol related circumsolar radiation is strongly
underestimated if IFS data are used head on, even for stable clear sky conditions. In this
section it is investigated if a statistical post-processing can be used to obtain reasonable
CSR values, though. This seemed promising, since it was already demonstrated that
the model fulfills two important prerequisites for a successful retrieval of the circumsolar
radiation: The model can predict the optical thickness (Bouarar et al., 2012) reasonably
well and furthermore can distinguish between periods at which large particles are present or
not (comp. Sect. 4.2.1). However, statistical post-processing comes at the cost of challenged
universal validity because it depends on reference measurements. It was exercised with the
measurements from the PSA and is not transferable one-to-one to other locations. A simple
coefficient fit was exercised nevertheless to explore if the model could in principle serve in
the retrieval of circumsolar radiation, specific tuning provided.
98 5. Discussion
Originally the CSR is calculated with the adding method (see Sect. 3.5.2) as:
CSR = 1.0− exp
Naerosol components∑i=1
−(τapp,sun,i − τapp,α,i)
, (5.1)
where τapp,sun,i and τapp,α,i are the apparent optical thickness values for the ith aerosol
component for a field of view enclosing the sun and for a field of view with the angular radius
α for which the CSR shall be calculated, respectively (compare also Eqs. 3.15 & 3.21).
Under the assumption of linearity one can also treat the contributions of the individual
aerosol components to the CSR separately (comp. Eq. 3.23):
CSR =
Naerosol components∑i=1
1.0− exp [−(τapp,sun,i − τapp,α,i)]
=
Naerosol components∑i=1
CSRi . (5.2)
The linearity assumption holds true for the considered IFS data for the location of the PSA.
The mean relative deviation between CSR calculated with Eq. (5.1) or with Eq. (5.2) is
only 0.2%. Therefore the latter equation was used to enable the separation of the individual
CSR-contributions.
In the post-processing step the CSR contributions of the individual aerosol types were
weighted by time independent coefficients, so that the post-processed CSR is given as
Naerosol components∑i=1
ci · CSRi . (5.3)
The coefficients were determined by minimizing the mean absolute deviation in CSR in
regard to the SAMS measurements using a truncated Newton algorithm1. It is an iterative
minimizing procedure and does not necessarily always find the global minimum as it would
be possible with an analytical linear least squares regression. For the purpose of this
exemplary evaluation this drawback is however not crucial. In fact, the utilized numerical
method is more flexible and contrasting the least squares method allows to use the mean
absolute deviation as cost function.
1Implemented in the Python software module scipy.optimize.fmin tnc (SciPy 0.13)
5.1 Statistical Post-Processing of IFS Aerosol Output 99
Several setups of the fitting process were examined. The filtered validation dataset –
containing the 1081 data tuples mentioned above in Sect. 4.2.1 – was always divided into a
training set and an evaluation set by varying criteria. To ensure to have best possible data
quality for the training, only measurements taken at θsun < 60 were used in this example
for the training.
If the coefficients for all aerosol components are fitted without constraints, the results are
ambiguous: Depending on how the training and validation set is chosen, or how the first
guess for the coefficients is made, the coefficients vary strongly, and sometimes extremely
large or small values are the result. Interpreting this, one should be aware of that the
contribution of most IFS aerosol components to the total CSR is negligible because they
are comprised of particles with radii much smaller than 1 µm and thus show no distinct
forward scattering peak. The algorithm therefore fits the measured CSR with several time
series of limited physical meaning, which may lead to the ambiguous results. The aerosol
component indices and the corresponding mean contribution to the CSR at PSA are listed
in Tab. 5.1. Furthermore some of the aerosol components are correlated which can cause
ambiguity in the fit results as well. While virtually all coefficients depend strongly on the
fitting setup, most of the tested setups have in common that they produce a coefficient c3
(for the large dust component) between 40 and 60. In the end one can force all coefficients
but c3 to be 1 without increasing the fit residual considerably.
In the following, the result of one specific fit setup is presented as example. Data from
the first nine days of each month were used for the training (191 data tuples) and data
from the remaining days for the evaluation (823 data tuples). All coefficients but c3 were
constrained to be 1. The fitted coefficient c3 in this case is 50.9, independent of the first
guess. With this coefficient the validation measures computed for the 823 data tuples in the
evaluation set improve considerably (comp. Tab. 5.2): The rank correlation is increased
from 0.33 to 0.70, the MAD is reduced from 0.010 to 0.005 and the bias is reduced from
-72% to -22%. The table also holds validation measures obtained by evaluating all 7972
data tuples originally considered (comp. Sect. 4.2.1) – including times with high temporal
variability in the measured CSR. If the coefficient c3 = 50.9 is applied on this data series,
the validation measures improve as well. However the results are still rather poor with a
bias of -49%, MAD of 0.013 and a rank correlation of 0.49. After the post-processing the
CSR signal is mainly determined by the dust large and the sea salt medium component.
Together they account for 81% of the CSR considering the evaluation dataset of 823 data
tuples. Originally the CSR is dominated by the sea salt medium aerosol component alone.
100 5. Discussion
If we again assume linearity, we can propagate the fit coefficients back and calculate the
AOT contributions of the individual aerosol components after the fitting. To this end, one
applies the same coefficients on the AOT as on the CSR. In the presented example the
linearity assumption introduces an error of few percent only. As mentioned before, the
largest particles are not necessarily accountable for large shares of the optical thickness.
This is reflected in the total aerosol optical thickness after applying the fit. The mean AOT
at 500 nm∗ for the period from 01 May. 2011 till 30 Apr. 2012 from the IFS data before
fitting is 0.11 and 0.15 after fitting. Considering the bias in regard to the AERONET value
(0.14) the fit also improves the AOT bias from -19% to +6%.
It was shown that a post-processing of the aerosol related CSR values can considerably
improve the results. However the exercised fit cannot increase the temporal resolution of
the model an thus not solve the problem with the sometimes high temporal variability
of the CSR. Furthermore the underrepresentation of large dust particles in the IFS, as
well as their importance, may vary with location. Therefore, the obtained results are not
applicable one-to-one to other locations. It was demonstrated, however, that weather model
output can in principle serve in the derivation of aerosol related circumsolar radiation,
although, in regard to the IFS, improvements concerning the aerosol size distribution are
necessary before the model output can be used unmodified. Future circumsolar radiation
measurements at different places on the globe could help to validate and improve weather
models in regard to large aerosol particles and/or to develop a more sophisticated time
and location dependent post-processing.
5.2 Uncertainties in the Cirrus Cloud Property Re-
trieval
Considering the calculation of cirrus related circumsolar radiation the cloud property re-
trieval poses a considerable uncertainty. In the following Sect. 5.2.1 especially the effect
of the utilized cirrus cloud mask on the calculated circumsolar radiation is discussed. In
Sect. 5.2.2 the reasons for spurious cloud property retrieval results at some pixels are
investigated.
∗Utilising 500 nm here as the closest wavelength to 550 nm supported by the AERONET sun photometerat the PSA.
5.2 Uncertainties in the Cirrus Cloud Property Retrieval 101
Table 5.1: Mean contribution to the CSR and AOT(500 nm) values of the individual aerosolcomponents for Almerıa derived from IFS data for the 823 data tuples of the filteredevaluation dataset before fitting.
i Aerosol Component CSR orig. AOT orig.1 Dust small 8.2e-4 7.6e-22 Dust medium 2.2e-4 8.1e-33 Dust large 1.4e-4 1.4e-34 Black carbon hydrophilic 1.1e-6 3.5e-45 Black carbon hydrophobic 2.3e-5 7.3e-36 Organic carbon hydrophilic 4.1e-6 6.7e-47 Organic carbon hydrophobic 9.8e-5 1.6e-28 Sea salt small 1.2e-4 1.0e-29 Sea salt medium 1.9e-3 5.7e-310 Sea salt large 6.0e-4 7.3e-411 Sulfate 2.7e-4 3.7e-2
Total 4.2e-3 0.164
Table 5.2: Validation measures computed for CSR as originally derived from IFS outputand for the fitted CSR time series with coefficient c3 = 50.9.
rrank,CSR rCSR MAD Bias NIFS original stable conditions 0.33 0.41 0.010 -72% 823IFS fitted (c3 = 50.9) stable conditions 0.70 0.75 0.005 -22% 823IFS original all data 0.25 0.17 0.015 -73% 7972IFS fitted (c3 = 50.9) all data 0.49 0.28 0.013 -49% 7972
5.2.1 Undetected Cirrus Clouds
As mentioned the APICS cloud property retrieval is only applied to pixels contained in a
cirrus cloud mask. In this section it is discussed how sensitive the circumsolar radiation
results are to the implemented change in the APICS cirrus cloud mask from MeCiDa to
COCS.
The detection efficiency of COCS is only better than MeCiDa’s for thin cirrus clouds
(Kox, 2012). Therefore, the total ice cloud cover (ICC), considering all evaluated SEVIRI
measurements in the test sector (May 2011 – Apr. 2012, θsun < 80), increases only
moderately from 17.4% to 22.2%, i.e. by 28%, when changing from MeCiDa to COCS
(see Tab. 5.3). However, if looking at measurements with a total irradiance Itot,αcir=2.5
larger than 200 W/m2 alone, the ICC determined by COCS is about 70%-80% larger than
102 5. Discussion
Table 5.3: Determined ice cloud cover (ICC) for different setups of APICS. The firstcolumn gives the total ice cloud cover. The second column holds the relative amount ofpixels classified as cirrus with Itot > 200 W/m2 (= Ncirrus,200 W/m2) on the evaluated slots(= Ntot) and the last column provides the relative amount of pixels classified as cirrus withItot > 200 W/m2 on all slots with Itot > 200 W/m2 (= N200 W/m2). Irradiance values werecalculated for αcir = 2.5.
APICS Setup ICC Ncirrus,200 W/m2/Ntot Ncirrus,200 W/m2/N200 W/m2
MeCiDa, Baum v2.0 17.4% 6.5% 8.8%COCS, Baum v2.0 22.2% 10.6% 14.0%MeCiDa, Baum v3.5 17.4% 5.8% 7.9%COCS, Baum v3.5 22.2% 9.9% 13.2%
Table 5.4: Mean circumsolar iradiance and mean CSR in case of cirrus clouds for differentsetups of APICS. All values were calculated for αcir = 2.5.
APICS Setup Icir(Itot > 200 W/m2) CSR for cirrus with (Itot > 200 W/m2)MeCiDa, Baum v2.0 5.4 W/m2 0.24COCS, Baum v2.0 7.8 W/m2 0.20MeCiDa, Baum v3.5 3.7 W/m2 0.19COCS, Baum v3.5 5.3 W/m2 0.15
the one from MeCiDa. The exact number depends on the considered ice particle shape
parameterization because it influences the calculation of Itot. In terms of mean circumsolar
irradiance this results in an increase of 40%-50% (considering an αcir of 2.5, see Tab. 5.4).
When MeCiDa is used, fewer thin clouds are detected, but the amount of thick clouds
remains basically unchanged. Therefore, the mean CSR in case of cirrus clouds decreases
in the order of 15% to 20% when changing from MeCiDa to COCS. Figure 5.1 shows
histograms of the CSR occurrence for the four setups of APICS as listet in Tab. 5.4 – that
is for the possible combinations of the COCS and MeCiDa cloud mask and the “Baum v2”
and “Baum v3.5” optical properties. The curves converge at large CSR values if different
cloud masks but the same cirrus optical properties are used. This confirms that the cirrus
detection schemes only differ for optically thin cirrus clouds. The discussed examples
highlight the importance of detecting and retrieving thin cirrus clouds at the best.
Even COCS does not detect all cirrus clouds, although it shows increased performance
5.2 Uncertainties in the Cirrus Cloud Property Retrieval 103
Figure 5.1: Histograms of the relative occurrence of CSR values due to cirrus cloudsunder the condition Itot,α=2.5 > 200W/m2 with respect to the total number of SEVIRImeasurements with Itot,α=2.5 > 200W/m2 for one year within the test sector consideringthe two ice particle shape parameterizations ”Baum v2.0” and ”Baum v3.5” as well as thetwo different cirrus cloud detection algorithms MeCiDa and COCS.
104 5. Discussion
compared to MeCiDa. In the following it is estimated how large the impact of the missed
cirrus clouds is in terms of circumsolar radiation. Bugliaro et al. (2012, Fig. 25.5 b) gave
the detection efficiency values ηdet for COCS and MeCiDa as step functions. For simplicity
they are approximated here as
ηdet,MeCiDa(τ) =
0.125 + 1.458 · τ, for τ < 0.6
1, else(5.4)
ηdet,COCS(τ) =
0.375 + 2.083 · τ, for τ < 0.3
1, else(5.5)
Figure 5.2 displays the approximated detection efficiencies. For each detected cloud, whose
optical thickness is retrieved by APICS, a correction factor for the occurrence can be
defined:
fcor(τ) =1
ηdet(τ). (5.6)
For example, if for a cloudy SEVIRI pixel fcor = 2 is identified, it is likely that there is
another pixel containing a cirrus with the same optical thickness τ that stays undetected.
Therefore, if computing an occurrence histogram, one can determine the average correction
factor for each histogram bin which allows to transform the real histogram by multiplication
by fcor into a new histogram corrected for the missed cloud detections. This is shown in
Fig. 5.3 were CSR histograms are displayed similar to Fig. 5.1 but also a version corrected
for the missed cloud detections (note that only SEVIRI measurements for the month May
2011 are the basis for Fig. 5.3). Furthermore a histogram is shown that was obtained by
treating all SEVIRI pixels as cirrus cloudy and processing them through the cloud property
retrieval. This will of course produce a lot of wrong values as water clouds, or even clear
sky measurements deviating from the APICS a priori assumptions will thereby produce
an ice optical thickness. However, this curve is shown as it poses the upper limit for the
histogram.
Interestingly the histogram correction mainly affects CSR values smaller than 0.2 for
MeCiDa and smaller than 0.1 for COCS. It can therefore not correct the discrepancy
between COCS and MeCiDa at higher CSR values. This, however, can be alleviated if the
maximum detection efficiency of MeCiDa is assumed to be 0.8 (histogram not shown). This
suggests that either MeCiDa reaches it full detection efficiency only at optical thickness
values considerably larger than 0.6 (comp. Eq. 5.5) or that COCS produces a considerable
amount of false alarms. However as mentioned earlier, Kox (2012) found a false alarm
5.2 Uncertainties in the Cirrus Cloud Property Retrieval 105
Figure 5.2: Cirrus detection efficiency for COCS and MeCiDa (approximating the stepfunction from Bugliaro et al., 2012, Fig. 25.5 (b)).
Figure 5.3: Histograms of the relative occurrence of CSR values due to cirrus cloudsunder the condition Itot,α=2.5 > 200W/m2 with respect to the total number of SEVIRImeasurements with Itot,α=2.5 > 200W/m2 for May 2011 within the test sector consideringthe ice particle shape parameterization ”Baum v2.0” as well as the cirrus cloud detectionalgorithms MeCiDa and COCS. The light colored lines show the results corrected fordecreasing cloud detection efficiency at lower optical thickness. The yellow curve wasobtained by assuming all SEVIRI pixels as cirrus cloud covered.
106 5. Discussion
rate of only approximately 5% for COCS using CALIOP spaceborne LIDAR observations
as reference, which makes the former point more likely. Also the data basis for τ > 0.6
from the LIDAR measurements evaluated by Bugliaro et al. (2012) is rather small so that
a concluding answer considering the detection efficiency at τ > 0.6 cannot be given. In
general it should be noted that Bugliaro et al. (2012) could rely only on about 450 data
points to calculate the detection efficiency distributions so that a considerable uncertainty
must be expected. It must also be noted that the histogram correction does not consider
the ice and water cloud cover in a specific scene. For example, if the ice cloud cover in a
hypothetical scene was 100%, fcor could still be larger than one which is unphysical. Still,
for a coarse estimate this method seems appropriate because the shown histograms are
calculated utilizing many different scenes.
Figure 5.4 shows the detection efficiency correction applied to histograms of circumsolar
irradiance (αcir = 2.5). The correction for MeCiDa affects Icir values smaller than ap-
proximately 100 W/m2 and the correction for COCS values smaller than approximately
75 W/m2. Again at higher values the discrepancy between COCS and MeCiDa stays un-
changed.
Figure 5.4: Same as Fig. 5.3 but here the occurrence of circumsolar irradiance values isdisplayed instead of occurrence of CSR values.
Concluding, it can be said that the variation of the curves in Figs. 5.3 & 5.4 (except for the
5.3 Comparison of the Satellite CSR Retrieval to a Regression Model 107
light green curves) depict the uncertainties remaining from the imperfect cloud detection
algorithms.
5.2.2 Spurious SEVIRI Pixels
There are some pixels in the SEVIRI image for which the average retrieved CSR values are
exceptionally high or low compared to the neighbouring pixels (comp. figures in Sect. 4.3).
This is because they show an exceptionally high or low amount of cloud detections. Two
reasons for this have been diagnosed. First, there are some pixels for which the COCS
algorithm spuriously diagnoses almost always clouds. Second, there are some pixels for
which often no albedo could be unraveled from the EUMETSAT Clear Sky Map (comp.
Sect. 3.2.2). This is because the Clear Sky Map often contains negative reflectivity values
for these pixels. In such case the APICS retrieval fails. In the processing of the satel-
lite measurements, the SEVIRI images were filtered through a land-water mask from the
LANDSAF1. Ocean pixels as well as continental pixels containing a considerable amount
of water are therefore masked. The reason for the negative values in the Clear Sky Map is
not entirely clear, but most pixels for which negative values occur are near spots that con-
tain water surfaces according to the LANDSAF land-water-mask. Possibly the land-water
mask does not always capture the full extent of the water surfaces. However the number
of spurious pixels is low and does not put the derived results into question.
5.3 Comparison of the Satellite CSR Retrieval to a
Regression Model
Neumann et al. (1998) parameterized the CSR as a function of the DNI/GHI ratio, because
they found that CSR cannot reasonably be parameterized from DNI alone. Combined DNI
and GHI measurements are taken at many sites so that their parameterization could in
principle help to alleviate the lack of CSR measurements. They published empirical fits to
their measurements of circumsolar radiation performed at the PSA and in Cologne. For
the PSA they parameterized the CSR in percent as
CSR = 70− 65.948 · DNI
GHI· (90 − θsun)0.669
15.822, (5.7)
1Land Surface Analysis Satellite Applications Facility.
108 5. Discussion
where the right most term normalizes the DNI/GHI ratio to empirically determined clear
sky conditions.
The SAMS dataset also contains DNI and GHI measured at the nearby meteorological
station as additional data. Therefore one can compare the fit (Eq. 5.7) to satellite retrieved
as well as to directly measured CSR. To allow for this comparison, CSR values for the PSA
calculated with Eq. (5.7) were re-gridded to the 15 min-interval of SEVIRI in the same
way like the SAMS measurements, as described in Sect. 4.2.2. The validation measures
listed in Tab. 5.5 indicate that the DNI/GHI-fitting method has advantages compared to
the satellite retrieval developed in this study if instantaneous CSR values shall be derived:
They show higher correlation and a lower mean absolute deviation when compared to the
SAMS values than the satellite derived values. This can be attributed to the fact that
DNI and GHI were measured at the same place as the CSR, and therefore all collocation
errors are removed, which need to be considered for SEVIRI measurements, though. The
statistical distribution of CSR, however, is better captured by the satellite retrieval. This
becomes obvious from the bias (Tab. 5.5): While the mean value from the DNI/GHI
parameterization is 50% off, the satellite derived values only show an underestimation of
11%. Furthermore the histogram for the satellite retrieval resembles the one for the SAMS
measurements considerably better than the one for the DNI/GHI-fit which can be followed
in Fig. 5.5. Therefore, I conclude that the, compared to the DNI/GHI-parameterization,
elaborate process of deriving the circumsolar radiation is rewarded with advantages in the
characterisation of the circumsolar radiation.
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8CSR
0
100
200
300
400
500
600
700
800
# N = 1890
APICS Baum 3.5
SAM Measurement
DNI/GHI param.
Figure 5.5: Histograms of CSR measured by SAMS and retrieved from SEVIRI (just asin Fig. 4.8), as well as parameterized according to Eq. 5.7 (red).
5.3 Comparison of the Satellite CSR Retrieval to a Regression Model 109
Table 5.5: Comparison of CSR retrieved from SEVIRI and obtained from the DNI/GHI-fit(Eq. 5.7) to values measured by SAMS: Rank correlation rrank,CSR, Pearson correlationrCSR, mean absolute deviation MAD, bias and the number of compared data tuples N .
Optical Properties rrank,CSR rCSR MAD Bias NAPICS, Baum v3.5 0.48 0.44 0.11 -11% 1890DNI/GHI-fit 0.86 0.86 0.09 50% 1890
It should be mentioned that Neumann et al. (1998) measured the DNI with an Eppley
Normal Incidence Pyrheliometer (NIP) which has a slightly larger field of view than the
Kipp & Zonen CH1/CHP1 that was used to measure the DNI values included in the
SAMS dataset. Also the CSR refers to a slightly different αcir. Therefore, a fit formula
derived from the SAMS dataset would probably possess slightly different coefficients. This
could explain the bias (Tab 5.5) to some extent, but not the deviations in the histogram:
Histograms for SAMS and SEVIRI have also been calculated for slightly larger field of
views. They are not shown here because they look similar to the one in Fig. 5.5.
110 5. Discussion
Chapter 6
Conclusions
In this study a novel method to determine circumsolar radiation from data on cirrus clouds
and aerosol was developed. The specific data sets used in the study are cirrus cloud
properties retrieved from MSG measurements and modelled aerosol data from ECMWF’s
IFS. Several components were linked together and improved where necessary to develop the
method. The Monte Carlo radiative transfer model MYSTIC was extended by introducing
a sun disk instead of a point source such that it can simulate the circumsolar region with
high accuracy. Furthermore, optical property data sets for the aerosol components as
simulated in the IFS were generated to enable radiative transfer simulations specifically
in regard to the IFS model output. With these tools at hand, it was possible to establish
a database of coefficients which allow the fast computation of circumsolar radiation by
simple analytical expressions from cloud or aerosol properties instead of performing time
consuming radiative transfer simulations.
While the IFS aerosol data could be processed through the circumsolar radiation param-
eterization as they were, the required cloud properties first had to be retrieved from the
MSG measurements. To obtain the required parameters cloud particle effective radius and
cloud optical thickness the APICS retrieval framework was employed. The success rate of
APICS was optimized by improving an important a priori assumption for the cloud prop-
erty retrieval – the reflectivity of the ground. A ground albedo dataset which is consistent
with the other a priori assumptions made in the cloud property retrieval algorithm was
created. It replaces the formerly used MODIS datasets. The APICS retrieval requires
a cirrus cloud mask to select which satellite pixels are processed. The within this study
introduced use of a cirrus cloud mask based on the COCS algorithm allows to consider
112 6. Conclusions
more optically thin clouds in the retrieval than originally possible with the MeCiDa cloud
mask. This increases the retrieved circumsolar irradiance on average by some 40% to 50%.
APICS was also extended with new lookup tables for individual ice particle shapes and
the “Baum v3.5” ice particle mixture to allow for an uncertainty analysis concerning the
particle shape.
6.1 Synopsis of Findings
Considering cirrus related circumsolar radiation, the uncertainties in the complete retrieval
chain due to necessary assumptions of the ice particle shape have been assessed to be large.
On average, the circumsolar irradiance derived relying on the “Baum v2” ice particle pa-
rameterization is some 50% larger than the one obtained using “Baum v3.5”. However,
the spatial distributions show similar patterns. The evaluated satellite data revealed that
the circumsolar radiation exhibits a considerable variability on the scale of a few SEVIRI
pixels which cannot be captured by ground measurements at few selected places. Despite
the mentioned differences between the two state-of-the-art ice particle parameterizations,
a validation of retrieved CSR values with ground measurements performed at the PSA,
Spain, showed good agreement for both ice particle mixtures. For this location the results
obtained using the “Baum” parameterizations differ only by some 15% to 25% regarding
the mean of the CSR, depending on choice of the considered time span. A specification of
the best particle mixture parameterization is ambiguous since none optimizes all validation
measures. Considering the complete evaluated time series, the CSR derived using “Baum
v2” exhibits a slightly smaller bias (4%) compared to the one obtained using “Baum v3.5”
(-11%). The mean absolute deviation is 0.11 for both variants. The performance of the
satellite retrieval of cirrus related CSR was also compared to a CSR parameterization
from Neumann et al. (1998) which uses ordinary DNI and GHI measurements as they are
collected at many sites: The satellite retrieved values show considerably less bias and a
histogram that much better resembles the SAMS measurements. The parameterization
from Neumann et al. (1998) correlates better with the SAMS measurements, though. This
comes to no surprise as the utilized DNI/GHI measurements were taken directly next to
the SAMS location so that no collocation errors can occur. Besides the determination of
spatially resolved circumsolar radiation characteristics, the developed method also allows
to calculate the overestimation of exploitable irradiation by a pyrheliometer due to cirrus
clouds. To provide an example, this was exercised for a hypothetical power plant exhibiting
6.2 Outlook 113
a lower operation limit of 200 W/m2 and an opening half-angle of 0.7. An average over-
estimation of the solar resource in the order of 0.5% due to cirrus clouds alone was found,
with the exact values depending on location. In the context of the overall uncertainties in
the resource assessment which can be in the order of 5% this may seem small. Yet, these
small values in the order of 0.5% translate into a profit overestimation of several hundred
thousand Euro per year for a typical 50 MWe solar thermal plant. At locations with stable
insolation conditions it is achievable to determine the long term solar resource with un-
certainties as low as 3%, at the same time there are some spots for which overestimations
of over 1% were retrieved in this thesis. Therefore in some cases a 1% bias could face a
3% random error. Consideration of the circumsolar radiation could therefore improve the
resource assessment considerably in such cases.
Concerning the aerosol related CSR, a strong underestimation by the method was diag-
nosed. Most likely this is mainly caused by an underrepresentation of large dust particles
(r ' 1 µm) in the IFS. Furthermore, circumsolar radiation fluctuations on the time scale
of minutes were found – even during clear sky conditions. These cannot be reproduced
with the current model resolution of the IFS. However, an a posteriori adjustment of the
contribution of large dust particles to the CSR improves all validation measures consider-
ably. On one hand this challenges the applicability of the model results at other locations
than the PSA for the time being because the necessary adjustment may vary with location.
On the other hand it shows that the model has some skill in predicting the course of the
aerosol concentrations, including large particles, as the adjustment comprises only a static
bias correction for the large dust component.
6.2 Outlook
Despite obvious uncertainties, a retrieval of cirrus related circumsolar radiation from meteo-
rological satellite observations can complement ground measurements since it offers several
unique advantages: Compared to ground measurements it is cheap. Satellite data are also
readily available for many regions of the world for longer time spans than any of the so
far available time series of ground measurements. Furthermore, the method allows to eas-
ily compare the circumsolar radiation characteristics of arbitrary sites as long as they are
within the field of view of the same satellite. The diagnosed small scale spatial variability
highlights that ground-based measurements of circumsolar radiation at a few sites alone
114 6. Conclusions
are not sufficient for a global characterisation. The further development of satellite based
cloud property retrievals can improve the circumsolar radiation products since the errors
in the retrieved cloud properties are a main source of uncertainty in circumsolar radiation.
In particular, information about the ice particle shape composition would help to reduce
the uncertainty. A manual screening of a subset of the validation data series considerably
improved the mean absolute deviation and correlation (MAD 0.07 instead of 0.09, rrank 0.76
instead of 0.65, “Baum v3.5”) which indicates that the presence of sub-scale water clouds
below or besides the cirrus compromises the satellite retrieved results. A detection and
appropriate treatment of “mixed cloud” pixels is therefore another open point for further
development.
The presented parameterization with its efficient lookup table approach can in principle be
extended and applied to other data sources as well. Regional aerosol models, for example,
with higher temporal and spatial resolution could help to overcome the current limitations
of the IFS. To this end an optical properties according to the new data source would be
required so that expanded k-tables can be calculated.
Bibliography
G. B. Airy. Transactions of the Cambridge Philosophical Society, volume 5, chapter Onthe Diffraction of an Object-glass with Circular Aperture, pages 283–291. 1835.
B. A. Baum, P. Yang, Andrew J. Heymsfield, Carl G. Schmitt, Yu Xie, Aaron Bansemer,Yong-Xiang Hu, and Zhibo Zhang. Improvements in Shortwave Bulk Scattering andAbsorption Models for the Remote Sensing of Ice Clouds. Journal of Applied Meteorologyand Climatology, 50:1037–1056, 2011. doi: 10.1175/2010JAMC2608.1.
B.A. Baum, A.J. Heymsfield, P. Yang, and S.T. Bedka. Bulk Scattering Models for theRemote Sensing of Ice Clouds. Part 1: Microphysical Data and Models. Journal OfApplied Meteorology, 44:1885–1895, 2005a.
B.A. Baum, P. Yang, A.J. Heymsfield, S. Platnick, M. D. King, Y. X. Hu, and S.T. Bedka.Bulk Scattering Models for the Remote Sensing of Ice Clouds. Part 2: NarrowbandModels. Journal Of Applied Meteorology, 44:1896–1911, 2005b.
A. Benedictow, A.M. Blechschmidt, I. Bouarar, E. Cuevas, H. Clark, J. Griesfeller, V. Hui-jnen, N. Huneeus, L. Jones, J. Kapsomenakis, S. Kinne, K. Lefever, M. Razinger,A. Richter, M. Schulz, W. Thomas, V. Thouret, M. Vrekoussis, A. Wagner, and C.Zerefos. Validation Report of the MACC Reanalysis of Global Atmospheric Composition:Period 2003-2010. KNMI, De Bilt, the Netherlands, 2012.
P. Blanc. International Solar Energy Society Webinar: Improving DNI Information for CSPDevelopment, 2013. URL http://www.ises-online.de/fileadmin/user_upload/
PDF/ISES-Webinar-DNI-May2013-MINES_ParisTech-Ed1.5.pdf. 29 May 2013, last ac-cessed 24 Jun 2013.
Max Born and Emil Wolf. Principles of Optics : Electromagnetic Theory of Propagation,Interference and Diffraction of Light. Cambridge Univ. Press, Cambridge, 7. (expanded)edition, 1999. ISBN 978-0-521-64222-4.
I. Bouarar, H. Clark, H. Flentje, V. Huijnen, L. Jones, J. Griesfeller, N. Huneeus,J. Kapsomenakis, S. Kinne, E. Katragkou, K. Lefever, M. Razinger, A. Richter, A.M.Blechschmidt, M. Schulz, A. Benedictow, W. Thomas, V. Thouret, M. Vrekoussis,A. Wagner, and C. Zerefos. Validation Report of the MACC Near Real Time Global
116 BIBLIOGRAPHY
Atmospheric Composition Service, System Evolution and Performance Statistics, Sta-tus to March 2012. KNMI, De Bilt, the Netherlands, 2012.
L. Bugliaro, T. Zinner, C. Keil, B. Mayer, R. Hollmann, M. Reuter, and W. Thomas.Validation of Cloud Property Retrievals with Simulated Satellite Radiances: A CaseStudy for SEVIRI. Atmos. Chem. Phys., 11(12):5603–5624, 2011. doi: 10.5194/acp-11-5603-2011.
Luca Bugliaro, Hermann Mannstein, and Stephan Kox. Ice Cloud Properties from Space.In Ulrich Schumann, editor, Atmospheric Physics: Background - Methods - Trends, pages417–432. Springer-Verlag, Berlin, 2012.
Luca Bugliaro, Andreas Ostler, Martin Wirth, Claudia Emde, and Ulrich Schumann. Val-idation of Cirrus Detection and Cirrus Optical Thickness Derived from MSG/SEVIRIwith an Airborne High Spectral Resolution Lidar. In preparation for Atmos. Meas. Tech.,2013.
D. Buie and A.G. Monger. The Effect of Circumsolar Radiation on a Solar ConcentratingSystem. Solar Energy, 76:181–185, 2004.
D. Buie, A.G. Monger, and C.J. Dey. Sunshape Distributions for Terrestrial Solar Simula-tions. Solar Energy, 74:113–122, 2003. doi: 10.1016/S0038-092X(03)00125-7.
R. Buras and B. Mayer. Efficient Unbiased Variance Reduction Techniques for MonteCarlo Simulations of Radiative Transfer in Cloudy Atmospheres: The Solution. Journalof Quantitative Spectroscopy and Radiative Transfer, 112:434–447, 2011. doi: 10.1016/j.jqsrt.2010.10.005.
R. Buras, T. Dowling, and C. Emde. New Secondary-Scattering Correction in DISORTwith Increased Efficiency for Forward Scattering. J. Quant. Spectrosc. Radiat. Transfer,112:2028–2034, 2011.
Julio Chaves. Introduction to Nonimaging Optics. CRC Press, Boca Raton, 2008.
CIMO Guide. Guide to Meteorological Instruments and Methods of Observation, 7thedition. Technical report, World Meteorological Organization, 2008.
J. G. DeVore, A. T. Stair, A. LePage, D. Rall, J. Atkinson, D. Villanucci, S. A. Rappaport,P. C. Joss, and R. A. McClatchey. Retrieving Properties of Thin Clouds from SolarAureole Measurements. J. Atmos. Oceanic Technol., 26:2531–2548, 2009. doi: 10.1175/2009JTECHA1289.1.
J. G. DeVore, J. A. Kristl, and S. A. Rappaport. Retrieving Cirrus Microphysical Propertiesfrom Stellar Aureoles. Journal of Geophysical Research: Atmospheres, 118:5679–5697,2013. doi: 10.1002/jgrd.50440.
BIBLIOGRAPHY 117
C. Emde and B. Mayer. Simulation of Solar Radiation During a Total Solar Eclipse:A Challenge for Radiative Transfer. Atmos. Chem. Phys., 7:2259–2270, 2007. doi:10.5194/acp-7-2259-2007.
C. Emde, R. Buras, and B. Mayer. ALIS: An Efficient Method to Compute High Spec-tral Resolution Polarized Solar Radiances Using the Monte Carlo Approach. J. Quant.Spectrosc. Radiat. Transfer, 112:1622–1631, 2011. doi: 0.1016/j.jqsrt.2011.03.018.
EUM MET REP 07 0132. Cloud Detection for MSG - Algorithm TheoreticalBasis Document. EUMETSAT, Darmstadt, Germany, v2 edition, Nov 2007.EUM/MET/REP/07/0132.
EUM OPS DOC 09 5164. Cloud Mask Factsheet. EUMETSAT, Darmstadt, Germany, v1edition, Oct 2010. EUM/OPS/DOC/09/5164.
EUM OPS DOC 09 5165. SEVIRI - Clear Sky Reflectance Map Factsheet. EUMETSAT,Darmstadt, Germany, v2b edition, Jul 2011. EUM/OPS/DOC/09/5165.
EUM MET TEN 12 0332. Conversion from Radiances to Reflectances for SEVIRIWarm Channels. EUMETSAT, Darmstadt, Germany, v4 edition, Oct 2012.EUM/MET/TEN/12/0332.
F. Ewald, L. Bugliaro, H. Mannstein, and B. Mayer. An Improved Cirrus DetectionAlgorithm MeCiDA2 for SEVIRI and its Validation with MODIS. Atmos. Meas. Tech.,6:309–322, 2013. doi: 10.5194/amt-6-309-2013.
Clemens Fricke, Andre Ehrlich, Manfred Wendisch, and Birger Bohn. Wiss. Mitteil. Inst. f.Meteorol. Univ. Leipzig, volume 50, chapter Infuence of Surface Albedo Inhomogeneitieson Remote Sensing of Optical Thin Cirrus Cloud Mikrophysics, pages 1–10. Inst. f.Meteorol. Univ. Leipzig, 2012.
Josef Konrad Gasteiger. Retrieval of Microphysical Properties of Desert Dust andVolcanic Ash Aerosols from Ground-based Remote Sensing. PhD thesis, Lud-wig–Maximilians–Universitat Munchen, 2011.
H. Grassl. Calculated Circumsolar Radiation as a Function of Aerosol Type, Field of View,Wavelength, and Optical Depth. Applied Optics, 10:2542–2543, 1971.
Donald Grether, David Evans, Arlon Hunt, and Michael Wahlig. Measurement and Anal-ysis of Circumsolar Radiation. In U.S. Department of Energy Insulation Resource As-sessment Program Review, Davis, CA, August 19-21, 1980.
W. Greuell and R.A. Roebeling. Towards a Standard Procedure for Validation of Satellite-Derived Cloud Liquid Water Path. Journal of Applied Meteorology and Climatology, 48:1575–1590, 2009. doi: 10.1175/2009JAMC2112.1.
118 BIBLIOGRAPHY
G. M. Hale and M. R. Querry. Optical Constants of Water in the 200-nm to 200-mmWavelength Region. Appl. Opt., 12:555–563, 1973. doi: 10.1364/AO.12.000555.
S.-H. Ham and B. J. Sohn. Assessment of the Calibration Performance of Satellite VisibleChannels Using Cloud Targets: Application to Meteosat-8/9 and MTSAT-1R. Atmos.Chem. Phys., 10:11131–11149, 2010. doi: 10.5194/acp-10-11131-2010.
J.E. Hansen and L.D. Travis. Light Scattering in Planetary Atmospheres. Space Sci. Rev.,16:527–610, 1974. doi: 10.1007/BF00168069.
M. Hess, P. Kopke, and I. Schult. Optical Properties of Aerosols and Clouds: The SoftwarePackage OPAC. Bulletin of the American Meteorological Society, 79:831–844, 1998.
E.J. Highwood. Suggested Refractive Indices and Aerosol Size Parameters for Use in Radia-tive Effect Calculations and Satellite Retrievals ADIENT / APPRAISE CP2 TechnicalReport, DRAFT V2. Technical report, Department of Meteorology, University of Read-ing, August 2009.
B. N. Holben, T. F. Eck, I. Slutsker, D. Tanre, J. P. Buis, A. Setzer, E. Vermote, J. A. Rea-gan, Y. J. Kaufman, T. Nakajima, F. Lavenu, I. Jankowiak, and A. Smirnov. AERONET– A Federated Instrument Network and Data Archive for Aerosol Characterization. Re-mote Sens. Environ., 66:1–16, 1998.
T. Holzer-Popp, M. Schroedter-Homscheidt, H. Breitkreuz, D. Martynenko, and L. Kluser.Improvements of Synergetic Aerosol Retrieval for ENVISAT. Atmospheric Chemistryand Physics, 8(24):7651–7672, 2008. doi: 10.5194/acp-8-7651-2008.
J. H. Joseph, W. J. Wiscombe, and J. A. Weinman. The Delta-Eddington Approximationfor Radiative Flux Transfer. J. Atmos. Sci., 33(12):2452–2459, December 1976. ISSN0022-4928. doi: 10.1175/1520-0469(1976)033〈2452:TDEAFR〉2.0.CO;2.
Ragini Kalapatapu, Matteo Chiesa, Peter Armstrong, and Stefan Wilbert. Measurementof DNI Angular Distribution with a Sunshape Profiling Irradiometer. In SolarPACES,Marrakesch, 09 2012.
K. Kandler, L. Schutz, C. Deutscher, M. Ebert, H. Hofmann, S. Jackel, R. Jaenicke,P. Knippertz, K. Lieke, A. Massling, A. Petzold, A. Schladitz, B. Weinzierl, A. Wieden-sohler, S. Zorn, and S. Weinbruch. Size Distribution, Mass Concentration, Chemi-cal and Mineralogical Composition and Derived Optical Parameters of the BoundaryLayer Aerosol at Tinfou, Morocco, During SAMUM 2006. Tellus B, 61(1), 2011. doi:10.1111/j.1600-0889.2008.00385.x.
S. Kato, T. P. Ackerman, J. H. Mather, and E. Clothiaux. The k-Distribution Methodand Correlated-k Approximation for a Shortwave Radiative Transfer Model. J. Quant.Spectrosc. Radiat. Transfer, 62:109–121, 1999.
BIBLIOGRAPHY 119
David L. King. Photovoltaic Module and Array Performance Characterization Methods forAll System Operating Conditions. AIP Conf. Proc., 394:347–368, 1996. doi: 10.1063/1.52852.
S. Kinne, T.P. Akermann, M. Shiobara, A. Uchiyama, A.J. Heymsfield, L. Miloshevich,J. Wendell, E.W. Eloranta, C. Purgold, and R.W. Bergstrom. Cirrus Cloud Radia-tive and Microphysical Properties from Ground Observations and In Situ MeasurementsDuring Fire 1991 and their Application to Exhibit Problems in Cirrus Solar RadiativeTransfer Modeling. Journal of the Atmospheric Sciences, 54:2320–2344, 1997.
S. Kox, A. Ostler, M. Vazquez-Navarro, L. Bugliaro, and H. Mannstein. Optical Prop-erties of Thin Cirrus Derived from the Infrared Channels of SEVIRI. In EUMETSATMeteorological Satellite Conference, Oslo, Norway, Darmstadt, 2011. EUMETSAT.
Stephan Kox. Remote Sensing of the Diurnal Cycle of Optically Thin Cirrus Clouds. PhDthesis, Ludwig–Maximilians–Universitat Munchen, 2012. URL http://nbn-resolving.
de/urn:nbn:de:bvb:19-151170.
W. Krebs, H. Mannstein, L. Bugliaro, and B. Mayer. Technical Note: A New Day- andNight-Time Meteosat Second Generation Cirrus Detection Algorithm MeCiDA. Atmos.Chem. Phys., 7:6145–6159, 2007.
Peter Kopke, Joachim Reuder, and Jan Schween. Spectral Variation of the SolarRadiation During an Eclipse. Meteorologische Zeitschrift, 10:179–186, 2001. doi:10.1127/0941-2948/2001/0010-0179.
Kuo-Nan Liou. An Introduction to Atmospheric Radiation. International Geophysics Series.26. Academic Press, New York, 1980. ISBN 0-12-451450-2.
N.G. Loeb and R. Davies. Angular Dependence of Observed Reflectances: A Comparisonwith Plane Parallel Theory. Journal of Geophysical Research, 102(D6):6865–6881, 1997.
N.G. Loeb, T. Varnai, and R. Davies. Effect of Cloud Inhomogeneities on the Solar ZenithAngle Dependence of Nadir Reflectance. Journal of Geophysical Research, 102(D8):9387–9395, 1997.
S. Lohmann, C. Schillings, B. Mayer, and R. Meyer. Long-Term Variability of Solar Directand Global Radiation Derived from ISCCP Data and Comparison with Reanalysis Data.Solar Energy, 80(11):1390–1401, November 2006. ISSN 0038-092X.
Kaj Madsen, Hans Bruun Nielsen, and Ole Tingleff. Methods for Non-Linear Least SquaresProblems. Informatics and Mathematical Modelling, Technical University of Denmark,DTU, Richard Petersens Plads, Building 321, DK-2800 Kgs. Lyngby, 2004. URL http:
//www2.imm.dtu.dk/pubdb/views/edoc_download.php/3215/pdf/imm3215.pdf. Lastaccessed 25 Jul. 2013.
120 BIBLIOGRAPHY
A. Mangold, H. De Backer, B. De Paepe, S. Dewitte, I. Chiapello, Y. Derimian, M. Kacene-lenbogen, J.-F. Leon, N. Huneeus, M. Schulz, D. Ceburnis, C. O’Dowd, H. Flentje,S. Kinne, A. Benedetti, J.-J. Morcrette, and O. Boucher. Aerosol Analysis and Fore-cast in the European Centre for Medium-Range Weather Forecasts Integrated ForecastSystem: 3. Evaluation by Means of Case Studies. Journal of Geophysical Research, 116(D3):D03302, February 2011. ISSN 0148-0227. doi: 10.1029/2010JD014864.
B. Mayer and A. Kylling. Technical Note: The libRadtran Software Package for RadiativeTransfer Calculations: Description and Examples of Use. Atmos. Chem. Phys., 5:1855–1877, 2005. doi: 10.5194/acp-5-1855-2005.
Bernhard Mayer. Radiative Transfer in the Cloudy Atmosphere. Eur. Phys. J. Conferences,1:75–99, 2009. doi: 10.1140/epjconf/e2009-00912-1.
G.M. McFarquhar and A.J. Heymsfield. The Definition and Significance of an EffectiveRadius for Ice Clouds. Journal of the Atmospheric Sciences, 55:2039–2052, 1998.
J. F. Meirink, R. A. Roebeling, and P. Stammes. Inter-Calibration of Polar Imager SolarChannels Using SEVIRI. Atmos. Meas. Tech. Discuss., 6(2):3215–3247, 2013. doi:10.5194/amtd-6-3215-2013.
Richard Meyer. International Solar Energy Society Webinar: DNI Informa-tion for CSP Development Part3: Improving Knowledge during Project Devel-opment, 2013. URL http://www.ises-online.de/fileadmin/user_upload/PDF/
ISES-Webinar-DNI-P3_RMeyer-Suntrace_20130529.pdf. 29 May 2013, last accessed14 Aug 2013.
Richard Meyer, Jesus Torres Butron, Gunnar Marquardt, Marko Schwandt, NorbertGeuder, Carsten Hoyer-Klick, Elke Lorenz, Annette Hammer, and Hans Georg Beyer.Combining Solar Irradiance Measurements and Various Satellite-Derived Products to aSite-Specific Best Estimate. In SolarPACES Symposium 2008, 2008.
G. Mie. Beitrage zur Optik truber Medien, speziell kolloidaler Metallosungen. Annalender Physik, Vierte Folge, 25(3):377–445, 1908.
Qilong Min and Minzheng Duan. Simultaneously Retrieving Cloud Optical Depth andEffective Radius for Optically Thin Clouds. Journal of Geophysical Research: Atmo-spheres, 110(D21):n/a–n/a, 2005. doi: 10.1029/2005JD006136.
J.-J. Morcrette, O. Boucher, L. Jones, D. Salmond, A. Beljaars P. Bechtold, A. Benedetti,A. Bonet, J. W. Kaiser, M. Razinger, M. Schulz, S. Serrar, A. J. Simmons, M. Sofiev,M. Suttie, A. M. Tompkins, and A. Untch. Aerosol Analysis and Forecast in the Euro-pean Centre for Medium-Range Weather Forecasts Integrated Forecast System: ForwardModeling. Journal of Geophysical Research, 114:D06206, 2009.
BIBLIOGRAPHY 121
Teruyuki Nakajima and Michael D. King. Determination of the Optical Thickness andEffective Particle Radius of Clouds from Reflected Solar Radiation Measurements. PartI: Theory. Journal of the Atmospheric Sciences, 47:1878–1893, 1990.
Teruyuki Nakajima, Masayuki Tanaka, and Toyotato Yamauchi. Retrieval of the OpticalProperties of Aerosols from Aureole and Extinction Data. Appl. Opt., 22(19):2951–2959,1983. doi: 10.1364/AO.22.002951.
A. Neumann and A. Witzke. The Influence of Sunshape on the DLR Solar Furnace Beam.Solar Energy, 66(6):447 – 457, 1999. ISSN 0038-092X. doi: 10.1016/S0038-092X(99)00048-1.
A. Neumann, B. von der Au, and P. Heller. Measurements of Circumsolar Radiation atthe Plataforma Solar (Spain) and in Cologne (Germany). In Proc. of Int. Solar EnergyConf. Solar Engineering. ASME, 1998.
Andreas Neumann, Andreas Witzke, Scott A. Jones, and Gregor Schmitt. RepresentativeTerrestrial Solar Brightness Profiles. Journal of Solar Energy Engineering Transactionsof the ASME, 124:198–204, 2002. doi: 10.1115/1.1464880.
J.E. Noring, D.F. Grether, and A.J. Hunt. Circumsolar Radiation Data: The LawrenceBerkely Laboratory Reduced Data Base. Technical Report NREL/TP-262-44292, Na-tional Renewable Energy Laboratory, 1991. URL http://rredc.nrel.gov/solar/
pubs/circumsolar/index.html. Last accessed 26 Jul. 2013.
Andreas Ostler. Validierung von optischen Wolkeneigenschaften aus MSG-Daten. Master-arbeit, Ludwig–Maximilians–Universitat Munchen, 2011.
Grant William Petty. A First Course in Atmospheric Radiation. Sundog Publishing,Madison, 2. edition, 2006. ISBN 0-9729033-1-3.
D. Pozo-Vazquez, S. Wilbert, C.A. Gueymard, L. Alados-Arboledas, F.J. Santos-Alamillos,and M.J. Granados-Munoz. Interannual Variability of Long Time Series of DNI and GHIat PSA, Spain. In SolarPaces 2011, 2011.
A. Rabl and P. Bendt. Effect of Circumsolar Radiation on Performance of Focusing Col-lectors. J. Sol. Energy Eng., 104:237, 1982.
Lord Rayleigh (J. W. Strutt). Phil. Mag., 41:107, 274 and 447, 1871.
Bernhard Reinhardt, Robert Buras, Luca Bugliaro, and Bernhard Mayer. Determinationof the Sunshape from Meteosat Second Generation. Technical report, 2012. SFERAProject Report, Deliverable R13.4.
Bernhard Reinhardt, Robert Buras, Luca Bugliaro, Stefan Wilbert, and Bernhard Mayer.Determination of Circumsolar Radiation from Meteosat Second Generation. Atmos.Meas. Tech. Discuss., 6:5835–5879, 2013. doi: 10.5194/amtd-6-5835-2013.
122 BIBLIOGRAPHY
P. Ricchiazzi, S. Yang, C. Gautier, and D. Sowle. SBDART: A Research and TeachingSoftware Tool for Plane-Parallel Radiative Transfer in the Earth’s Atmosphere. Bulletinof the American Meteorological Society, 79:2101–2114, 1998.
J. M. Ritter and K. J. Voss. A New Instrument for Measurement of the Solar AureoleRadiance Distribution from Unstable Platforms. Journal of Atmospheric and OceanicTechnology, 17:1040–1047, 2000.
Helmut Scheffler and Hans Elsasser. Physik der Sterne und der Sonne, 2. Auflage. BI-Wiss.-Verl., Mannheim, 1990.
J. Schmetz, P. Pili, S. Tjemkes, D. Just, J. Kerkmann, S. Rota, and A. Ratier. An Intro-duction to Meteosat Second Generation (MSG). Bulletin of the American MeteorologicalSociety, 83:977–992, 2002.
M. Schubnell. Influence of Circumsolar Radiation on Aperture, Operating Temperatureand Efficiency of a Solar Cavity Receiver. Solar Energy Materials and Solar Cells, 27(3):233 – 242, 1992. ISSN 0927-0248. doi: 10.1016/0927-0248(92)90085-4.
Gregory L. Schuster, Oleg Dubovik, and Brent N. Holben. Angstrom Exponent and Bi-modal Aerosol Size Distributions. Journal of Geophysical Research: Atmospheres, 111(D7), 2006. ISSN 2156-2202. doi: 10.1029/2005JD006328.
Michal Segal-Rosenheimer, Phillip B. Russell, John M. Livingston, S. Ramachandran, JensRedemann, and Bryan A. Baum. Retrieval of Cirrus Properties by Sunphotometry: ANew Perspective on an Old Issue. Journal of Geophysical Research: Atmospheres, 118:1–18, 2013. ISSN 2169-8996. doi: 10.1002/jgrd.50185.
E.P. Shettle and R.W. Fenn. Models for the Aerosols of the Lower Atmosphere and theEffects of Humidty Variations on their Optical Properties. Technical Report AFGL-TR-79-0214, Air Force Geophysics Laboratory, Hanscom AFB, Massachusetts, 1979.
M. Shiobara and S. Asano. Estimation of Cirrus Optical Thickness from Sun PhotometerMeasurements. Journal of Applied Meteorology, 33(6):672–681, 1994. doi: 10.1175/1520-0450(1994)033〈0672:EOCOTF〉2.0.CO;2.
K. Stamnes, S.C. Tsay, W. Wiscombe, and K. Jayaweera. A Numerically Stable Algorithmfor Discrete-Ordinate-Method Radiative Transfer in Multiple Scattering and EmittingLayered Media. Applied Optics, 27(12):2502–2509, 1988.
K. Stamnes, S.-C. Tsay, W. Wiscombe, and I. Laszlo. DISORT, a General-Purpose FortranProgram for Discrete-Ordinate-Method Radiative Transfer in Scattering and EmittingLayered Media: Documentation of Methodology. Technical report, Dept. of Physics andEngineering Physics, Stevens Institute of Technology, Hoboken, NJ 07030, 2000.
BIBLIOGRAPHY 123
Olaf Stein, Martin Schultz, Johannes Flemming, Antje Inness, Johannes Kaiser, LukeJones, Angela Benedetti, and Jean-Jacques Morcrette. MACC Global Air QualityServices – Technical Documentation. Technical report, Forschungszentrum Julich /ECMWF, Julich / Reading, November 2011.
A.H. Strahler, J.P. Muller, and MODIS Science Team Members. MODIS BRDF/Albedoproduct: Algorithm Theoretical Basis Document Version 5.0, 1999. URL http://modis.
gsfc.nasa.gov/data/atbd/atbd_mod09.pdf. Last accessed 22 Mar. 2011.
Edith Thomalla, Peter Kopke, Heinz Mueller, and Heinrich Quenzel. Circumsolar Radi-ation Calculated for Various Atmospheric Conditions. Solar Energy, 30:575–587, 1983.doi: 10.1016/0038-092X(83)90069-5.
M. Waldmeier. Ergebnisse und Probleme der Sonnenforschung - Probleme der kosmischenPhysik, volume 22. Akademische Verlagsgesellschaft Becker und Erler, Leipzig, 1941.
Kathrin Wapler. Der Einfluss des dreidimensionalen Strahlungstransportes auf Wolken-bildung und -entwicklung. PhD thesis, Ludwig-Maximilians-Universitat Munchen, 2007.URL http://nbn-resolving.de/urn:nbn:de:bvb:19-68948.
A. D. Watt. Circumsolar Radiation. Technical report, Sandia National Laboratories,Albuquerque, New Mexico, 1980.
Viktor Wesselak and Thomas Schabbach. Regenerative Energietechnik. Springer-Verlag,Berlin, Heidelberg, 2009. doi: 10.1007/978-3-540-95882-6.
M. Wiegner, J. Gasteiger, K. Kandler, B. Weinzierl, K. Rasp, M. Esselborn, V. Freuden-thaler, B. Heese, C. Toledano, M. Tesche, and D. Althausen. Numerical Simulationsof Optical Properties of Saharan Dust Aerosols with Emphasis on Lidar Applications.Tellus B, 61(1), 2011. doi: 10.1111/j.1600-0889.2008.00381.x.
S. Wilbert, B. Reinhardt, J. DeVore, M. Roger, R. Pitz-Paal, and C. Gueymard. Measure-ment of Solar Radiance Profiles with the Sun and Aureole Measurement System (SAM).In SolarPaces 2011, 2011.
S. Wilbert, B. Reinhardt, J. DeVore, M. Roger, R. Pitz-Paal, C. Gueymard, and R. Buras.Measurement of Solar Radiance Profiles With the Sun and Aureole Measurement System(SAM). Journal of Solar Energy Engineering, 135:041002, 2013. doi: 10.1115/1.4024244.
Stefan Wilbert, Robert Pitz-Paal, and Joachim Jaus. Circumsolar Radiation and BeamIrradiance Measurements for Focusing Collectors. In ES1002: Cost Wire Workshop,Riso, 2012. URL http://elib.dlr.de/78602/.
D.S. Wilks. Statistical Methods in the Atmospheric Sciences. Academic Press Inc., Ams-terdam, 2 edition, 2005.
124 BIBLIOGRAPHY
P. Yang, K.N. Liou, K. Wyser, and D. Mitchell. Parameterization of the Scattering andAbsorption Properties of Individual Ice Crystals. Journal of Geophysical Research, 105(D4):4699–4718, 2000.
P. Yang, H. Wei, H.-L. Huang, B.A. Baum, Y.X. Hu, G.W. Kattawar, M.I. Mishchenko,and Q. Fu. Scattering and Absorption Property Database for Nonspherical Ice Particlesin the Near- Through Far-Infrared Spectral Region. Applied Optics, 44(26):5512–5523,2005.
Wilford Zdunkowski, Thomas Trautmann, and Andreas Bott. Radiation in the Atmosphere:A Course in Theoretical Meteorology. Cambridge University Press, Cambridge, 2007.ISBN 978-0-521-87107-5.
Glossary
AOT . . . . . . . . . . . . . . . Aerosol optical thickness
APICS . . . . . . . . . . . . . . Algorithm for the Physical Investigation of Clouds
BC . . . . . . . . . . . . . . . . Black carbon
COCS . . . . . . . . . . . . . . . Cloud Optical properties derived from CALIOP andSEVIRI algorithm
CSR . . . . . . . . . . . . . . . . Circum solar ratio
CST . . . . . . . . . . . . . . . . Concentrating solar technology
DNI . . . . . . . . . . . . . . . . Direct normal irradiance
ECMWF . . . . . . . . . . . . . European Centre for Medium-Range Weather Forecasts
EUMETSAT . . . . . . . . . . European Organisation for the Exploitation of Meteoro-logical Satellites
FAR . . . . . . . . . . . . . . . . False alarm rate
GHI . . . . . . . . . . . . . . . . Global horizontal irradiance
ICC . . . . . . . . . . . . . . . . Ice cloud cover
IFS . . . . . . . . . . . . . . . . Integrated forecast system
INSO . . . . . . . . . . . . . . . Insoluble
LANDSAF . . . . . . . . . . . . Land surface analysis satellite applications facility
LE . . . . . . . . . . . . . . . . . Local estimate
LIDAR . . . . . . . . . . . . . . Light detection and ranging
126 Glossary
LUT . . . . . . . . . . . . . . . . Lookup Table
MeCiDa . . . . . . . . . . . . . Meteosat Second Generation Cirrus Detection Algorithm
MIAM . . . . . . . . . . . . . . Mineral accumulation mode
MICM . . . . . . . . . . . . . . Mineral coarse mode
MINM . . . . . . . . . . . . . . Mineral nucleation mode
MODIS . . . . . . . . . . . . . . Moderate-Resolution Imaging Spectroradiometer (aboardNASA’s polar orbiting satellites Terra and Aqua)
MSG . . . . . . . . . . . . . . . Meteosat Second Generation meteorological satellite se-ries
MYSTIC . . . . . . . . . . . . . Monte Carlo Code for the Physically Correct Tracing ofPhotons in Cloudy Atmospheres
NRT . . . . . . . . . . . . . . . Near real time
OC . . . . . . . . . . . . . . . . Organic carbon
OPAC . . . . . . . . . . . . . . Optical Properties of Aerosols and Clouds
PSA . . . . . . . . . . . . . . . . Plataforma Solar de Almerıa
RH . . . . . . . . . . . . . . . . Relative humidity
RT . . . . . . . . . . . . . . . . . Radiative transfer
RTE . . . . . . . . . . . . . . . . Radiative transfer equation
SAM . . . . . . . . . . . . . . . Sun and Aureole Measurement System
SAMS . . . . . . . . . . . . . . . SAM based system
SEVIRI . . . . . . . . . . . . . . Spinning Enhanced Visible Infra-Red Imager (aboard theMSG satellites)
SSAM . . . . . . . . . . . . . . . Sea salt accumulation mode
SSCM . . . . . . . . . . . . . . . Sea salt coarse mode
TOA . . . . . . . . . . . . . . . Top of atmosphere
WASO . . . . . . . . . . . . . . Water-soluble
Acknowledgements
First of all I would like to thank my advisors Robert Buras and Bernhard Mayer. Theywere always supportive when I had problems or concerns and shared their ideas with me –many who have found their way into this document. They also provided linguistic supportin stylisation of the manuscript. Luca Bugliaro did the predominant programming workconcerning the improvement of APICS. He also developed the software code to correct forthe parallax in the SEVIRI measurements which I used in the validation section. Manyideas concerning the improvement of APICS came from his side. I would like to thank himfor many fruitful discussions. Josef Gasteiger and Claudia Emde provided useful insightsconcerning optical properties of aerosols and clouds. Josef’s database of pre-calculatedaerosol optical properties enabled the generation of bulk aerosol optical properties forlibRadtran resembling the physical properties of the aerosol as in the ECMWF IFS. Iwould like to thank Peter Kopke for his advice on OPAC and aerosols in general. Specialthanks also go to Stephan Kox who provided me the COCS data and performed severalprocessing runs specifically for me. Stefan Wilbert provided ground measurements of thecircumsolar radiation as well as other meteorological measurands at the Plataforma Solarde Almerıa. I would also like to thank him for our always inspiring exchange of ideas aswell as for the hospitality that I received during my 2-months stay with DLR-Almerıa.Many thanks also to my other colleagues at DLR and MIM for the good times and thesupportive atmosphere – especially to my office mate Fabian. Last but not least I wouldlike to thank my family, especially my parents Martina and Wolfgang and my girl friendChristina, for always backing me up.
This study was mainly founded by the EU’s FP7 through the project “Solar Facilities forthe European Research Area – SFERA”.